Thesis Used Openfoam

University of Tennessee, Knoxville
Trace: Tennessee Research and Creative

Exchange
Masters Teses Graduate School
5-2012
Development and Verifcation of a Navier-Stokes
Solver with Vorticity Confnement Using
OpenFOAM
Austin Barret Kimbrell
University of Tennessee, austin.b.kimbrell@gmail.com
Tis Tesis is brought to you for free and open access by the Graduate School at Trace: Tennessee Research and Creative Exchange. It has been
accepted for inclusion in Masters Teses by an authorized administrator of Trace: Tennessee Research and Creative Exchange. For more information,
please contact trace@utk.edu.
Recommended Citation
Kimbrell, Austin Barret, "Development and Verifcation of a Navier-Stokes Solver with Vorticity Confnement Using OpenFOAM. "
Master's Tesis, University of Tennessee, 2012.
htp://trace.tennessee.edu/utk_gradthes/1173
To the Graduate Council:
I am submiting herewith a thesis writen by Austin Barret Kimbrell entitled "Development and
Verifcation of a Navier-Stokes Solver with Vorticity Confnement Using OpenFOAM." I have examined
the fnal electronic copy of this thesis for form and content and recommend that it be accepted in partial
fulfllment of the requirements for the degree of Master of Science, with a major in Mechanical
Engineering.
John S. Steinhof, Major Professor
We have read this thesis and recommend its acceptance:
K. C. Reddy, Joseph C. Yen
Accepted for the Council:
Carolyn R. Hodges
Vice Provost and Dean of the Graduate School
(Original signatures are on fle with ofcial student records.)
Development and Verification of a Navier-Stokes
Solver with Vorticity Confinement Using OpenFOAM

A Thesis Presented for the
Master of Science
Degree
The University of Tennessee, Knoxville

Austin Barrett Kimbrell
May 2012

ii

Copyright 2012 by Austin B. Kimbrell
All rights reserved.

iii
ACKNOWLEDGEMENTS

I would like to express sincere appreciation to my advisor Dr. John
Steinhoff for his guidance in developing this work and providing a challenging,
yet enjoyable experience. I would also like to thank my other advising committee
members, Dr. K.C. Reddy and Dr. Joseph Yen, for their time and helpful
comments. I would like to especially acknowledge Dr. Yen for all of the
discussion, insight, and advice he has provided over the past few years on topics
both work and school-related, which has helped grow my technical knowledge
and expertise in computational methods and aerodynamics. He is not only a
friend but a true mentor and I am very grateful to work with him.
I offer special thanks and acknowledgement to Chris Connor and Dr. Ed
Duell at Jacobs Technology for all of the support and opportunities for technical
advancement given to me. Jacobs Technology has been an invaluable resource
the past few years by providing funding for my continued education as well as the
available computational resources to complete this work. The working
environment at Jacobs is truly unique and I am thankful to be a part of it.
Lastly, I owe the greatest thanks to my wife, Oxana, and my daughter,
Sophia, for all of their support and understanding during this whole endeavor. It
wasnt always easy, but they were always there for me to provide
encouragement and keep me on the path to success.

iv
ABSTRACT

Vorticity Confinement (VC) is a numerical technique which enhances
computation of fluid flows by acting as negative diffusion within the limit of
vortical regions, preventing the inherent numerical dissipation present with
conventional methods. VC shares similarities with large eddy simulation (LES),
but its behavior is based on a stable negative dissipation of vortical structures
controlled by the automatic balance between two parameters, [mu] and
[epsilon].
In this thesis, three-dimensional, parallel-computing Navier-Stokes solvers
with VC are developed using the OpenFOAM computational framework.
OpenFOAM is an open-source collection of C++ libraries developed for
computational fluid dynamics. Object-oriented programming concepts are used to
develop the finite volume solvers, which introduce the VC source term into the
governing equations as a body force. An immersed boundary method is
implemented with the VC module to mitigate limitations of body-fitted grids.
The developed solvers are examined using two-dimensional boundary
layer simulations, which demonstrate that for a given range of confinement
parameters the boundary layer can be relaxed to a desired height to approximate
a turbulent boundary layer thickness. Unlike wall function models, however, the
VC boundary layer can still separate in an adverse pressure gradient. The
application of VC to a two-dimensional advecting compact vortex results in the
propagation of the vortex without dissipation. Solutions for a three-dimensional
backward-facing step are validated against experimental data. The VC
simulations show excellent agreement with experimental data for a fixed value of
[mu] and a given range of [epsilon]. Coarsening the mesh increases inherent
numerical dissipation and requires using a smaller value of [mu] to show good
agreement with experimental data. Turbulence kinetic energy spectra exhibit a
-5/3 slope inertial wavenumber range indicating proper turbulence cascading
using the VC model. Simulation of a Formula One racecar represented by an

v
immersed surface demonstrates the suitability of VC for fast prototyping. A time-
accurate VC analysis on a 3,400,000 cell coarse mesh appears more realistic in
the wake region than a steady RANS simulation on a 30,000,000 cell mesh. The
VC solution appears visually comparable to an LES solution but represents a
fraction of the computational cost.

vi
TABLE OF CONTENTS

Chapter 1 Introduction .......................................................................................... 1
1.1 Objectives ................................................................................................... 3
1.2 General Approach ....................................................................................... 4
Chapter 2 Vorticity Confinement ........................................................................... 6
2.1 VC1 Formulation ......................................................................................... 7
2.1.1 Surface Confinement ............................................................................ 9
2.2 VC2 Formulation ....................................................................................... 10
2.3 Compressible Vorticity Confinement ......................................................... 11
2.4 Dimensional Analysis of .......................................................................... 13
2.4.1 Dynamic Vorticity Confinement ........................................................... 15
Chapter 3 Introduction to OpenFOAM ................................................................ 16
3.1 Introduction to Object-Oriented Programming ........................................... 17
3.2 Programming in OpenFOAM ..................................................................... 19
3.3 OpenFOAM Data Structure ....................................................................... 20
3.4 Standard Case Structure ........................................................................... 21
3.4.1 system Directory ................................................................................. 24
3.4.2 constant Directory ............................................................................... 24
3.4.3 Time Directories .................................................................................. 26
3.5 OpenFOAM Utilities ................................................................................... 26
3.5.1 Pre-processing .................................................................................... 27
3.5.2 Meshing .............................................................................................. 27
3.5.3 Domain Decomposition ....................................................................... 28
3.5.4 Post-processing .................................................................................. 29
3.5.5 Run-time Processing........................................................................... 30
3.5.6 Solution Monitoring ............................................................................. 30
3.5.7 Additional Community-developed Tools.............................................. 31
Chapter 4 Solution Methodology ......................................................................... 32
4.1 Finite Volume Approach ............................................................................ 33

vii
4.1.1 Conservation Equations in Integral Form ............................................ 34
4.2 Pressure-Correction Methods ................................................................... 37
4.2.1 PISO Algorithm for Time-Accurate Simulation .................................... 41
4.2.2 SIMPLE Algorithm for Steady-State Simulation .................................. 43
4.2.3 Rhie-Chow Treatment in OpenFOAM ................................................. 44
4.3 Vorticity Confinement Implementation ....................................................... 45
4.3.1 Parallelization ..................................................................................... 47
4.3.2 VC as an Immersed Boundary Method ............................................... 49
Chapter 5 Results and Discussion ...................................................................... 51
5.1 Flow over a Flat Plate ................................................................................ 52
5.1.1 Setup Details ...................................................................................... 52
5.1.2 Baseline Results ................................................................................. 54
5.1.3 Effect of VC on Boundary Layer ......................................................... 56
5.2 Flow over a Slanted Plate ......................................................................... 62
5.2.1 Setup Details ...................................................................................... 63
5.2.2 Immersed Boundary with VC1 ............................................................ 64
5.2.3 Immersed Boundary with VC2 ............................................................ 67
5.3 Advection of a Compact Vortex ................................................................. 70
5.3.1 Setup Details ...................................................................................... 70
5.3.2 Numerical Stability Requirements ....................................................... 72
5.3.3 Simulation Results .............................................................................. 73
5.4 Flow over a Backward-facing Step ............................................................ 79
5.4.1 Setup Details ...................................................................................... 80
5.4.2 Mean Flow Comparison ...................................................................... 81
5.4.3 Instantaneous Flow Comparison ........................................................ 97
5.4.4 Turbulence Kinetic Energy Spectra .................................................. 104
5.4.5 Coarsened Mesh Solutions ............................................................... 108
5.5 Complex Immersed Body Simulation ...................................................... 118
5.5.1 Time-Accurate VC Study .................................................................. 119
5.5.2 High Resolution RANS Study ........................................................... 121

viii
5.5.3 Vorticity Field Comparison ................................................................ 124
5.5.4 Streamline Comparison .................................................................... 127
Chapter 6 Conclusion ....................................................................................... 134
6.1 Future Work ............................................................................................ 137
Bibliography ...................................................................................................... 139
Appendix ........................................................................................................... 148
Vita .................................................................................................................... 154

ix
LIST OF TABLES

Table 1. Simulation parameters for initial simulations in 5.4. .............................. 82

x
LIST OF FIGURES

Figure 1. OpenFOAM C++ function call identical to Equation (3.1). ................... 20
Figure 2. Two cell polyhedral mesh [45]. ............................................................ 22
Figure 3. OpenFOAM linked list mesh representation of two cell mesh in Figure 2
[45]. .............................................................................................................. 22
Figure 4. OpenFOAM case directory structure. .................................................. 23
Figure 5. Flow chart comparison of PISO and SIMPLE algorithms. .................... 42
Figure 6. OpenFOAM C++ code snippet for the PISO algorithm. ....................... 45
Figure 7. C++ code snippet for harmonic mean algorithm. ................................. 48
Figure 8. Computational mesh for flat plate case study. Processor subdomains
have been separately colored to show domain decomposition. ................... 53
Figure 9. Boundary conditions for flat plate case study. ...................................... 53
Figure 10. Velocity magnitude contour for flat plate case without VC. ................ 55
Figure 11. Vorticity magnitude contour for flat plate case without VC. ................ 55
Figure 12. Boundary layer profile of flat plate case without VC at various
locations along the plate. (x) refers to the distance from the leading edge. . 56
Figure 13. Varying VC1 confinement strength and its effect on boundary layer
height at downstream station x = 1.875 m. .................................................. 58
Figure 14. Varying VC2 confinement strength and its effect on boundary layer
height at downstream station x = 1.875 m. .................................................. 58
Figure 15. Boundary layer profile from the flat plate simulation with VC1
confinement strength =0.25. ....................................................................... 59
Figure 16. Boundary layer profile from the flat plate simulation with VC2
confinement strength =0.004. ..................................................................... 59
Figure 17. Velocity magnitude contours for VC1 (top) and VC2 (bottom)
simulations with confinement strength values of 0.25 (VC1) and 0.004 (VC2).
Note that a constant diffusion () of 0.001 is applied. .................................. 60
Figure 18. Vorticity magnitude contours for VC1 (top) and VC2 (bottom)
simulations with confinement strength values of 0.25 (VC1) and 0.004 (VC2).
Note that a constant diffusion () of 0.001 is applied. .................................. 61
Figure 19. Boundary conditions for slanted plate case study. Immersed body is
shown superimposed on the mesh. ............................................................. 63
Figure 20. Velocity magnitude contour of immersed plate without VC. ............... 65
Figure 21. Vorticity magnitude contour of immersed plate without VC. ............... 65

xi
Figure 22. Velocity magnitude contour for immersed plate with VC1. ................. 66
Figure 23. Vorticity magnitude contour for immersed plate with VC1. ................. 66
Figure 24. Velocity magnitude contour for immersed plate with VC2 and
confinement strength of 0.004. The boundary layer is exhibiting roll-up
behavior. ...................................................................................................... 68
Figure 25. Vorticity magnitude contour for immersed plate with VC2 and
confinement strength of 0.004. The boundary layer is exhibiting roll-up
behavior. ...................................................................................................... 68
Figure 26. Velocity magnitude contour for immersed plate with VC2 and first-
order upwind differencing for the convection term. The added numerical
diffusion causes the boundary layer to thicken and suppresses the transient
instabilities. .................................................................................................. 69
Figure 27. Velocity magnitude contour for immersed plate with VC2 and second-
order upwind differencing for the convection term, but with a reduced
confinement strength of 0.0035. Boundary layer instabilities are no longer
present in the solution. ................................................................................. 69
Figure 28. Boundary conditions for compact vortex advection case study. ......... 71
Figure 29. Centerline vorticity magnitude after 1 s (100 time steps). Without VC
the vortex strength has decreased an order of magnitude. .......................... 73
Figure 30. Contour plots of vorticity magnitude at one second simulation duration
(100 time steps). Note reduction of scale for simulation without VC. ........... 74
Figure 31. Centerline vorticity magnitude after 10 s (1000 time steps). ............. 75
Figure 32. Vorticity magnitude contour after 10 s (1000 time steps). Baseline
results not shown because the vortex has been absorbed into the
background flow. .......................................................................................... 76
Figure 33. Centerline vorticity magnitude after 40 s (4000 time steps). .............. 77
Figure 34. Vorticity magnitude contour after 40 s (4000 time steps). Low
magnitude vorticity concentrations are visible in VC1 solution. .................... 77
Figure 35. Schematic representation of backward-facing step geometry (h =
0.0254 m). .................................................................................................... 80
Figure 36. LES-typical non-uniform mesh used for initial VC studies in 5.4. Image
depicts the 248 x 57 cell mesh in the X and Y coordinate directions; 20
uniform grid cells are used in the Z coordinate direction (normal to plane). . 81
Figure 37. Iso-volume of mean streamwise velocity showing region of U < 0.0 for
the laminar solution (no VC). ....................................................................... 84
Figure 38. Mean streamwise velocity contour for laminar solution (no VC). ....... 85
Figure 39. Mean streamwise velocity contour for VC1 solution. Contour levels
from Figure 38 apply. ................................................................................... 85

xii
Figure 40. Mean streamwise velocity profiles from VC1 solution compared to
experimental data, LES results, and the laminar solution. Y-axis position has
been normalized by step height, while velocity has been normalized by the
prescribed inlet velocity. ............................................................................... 86
Figure 41. Streamwise RMS velocity profiles from the VC1 solution compared to
experimental data, LES results, and the laminar solution. ........................... 87
Figure 42. Mean streamwise velocity contours for VC2 solutions; = 1.6 (top),
= 2.0 (middle), = 2.5 (bottom). ............................................................... 89
Figure 43. Mean streamwise velocity profiles from the VC2 = 1.6 solution
compared to experimental data, LES results, and the laminar solution. ...... 91
Figure 44. Streamwise RMS velocity profiles from the VC2 = 1.6 solution
Figure 49. Horizontal section slices at Y = 0 height, level with the step leading
edge. ............................................................................................................ 98
Figure 50. Instantaneous center plane slices of vorticity magnitude, from
VISUALIZATION 1, laminar solution (top) and VISUALIZATION 3, VC1
solution (bottom). ......................................................................................... 99
Figure 51. Instantaneous volumetric-rendered images of vorticity magnitude,
from VISUALIZATION 2, laminar solution (top) and VISUALIZATION 4, VC1
solution (bottom). Contour levels below 2000 are invisible to accentuate
large coherent structures. Upper wall solution has been clipped to
emphasize shear layer structures. ............................................................. 100
Figure 52. Instantaneous center plane slices of vorticity magnitude, from
VISUALIZATION 5 with VC2 = 1.6 (top), VISUALIZATION 7 with VC2 =
2.0 (middle), and VISUALIZATION 9 with VC2 = 2.5 (bottom). ........... 102
Figure 53. Instantaneous volumetric-rendered images of vorticity magnitude,
from VISUALIZATION 2, baseline laminar solution (top left),
VISUALIZATION 6, VC2 = 1.6 (top right), VISUALIZATION 8, VC2 =
2.0 (bottom left), and VISUALIZATION 10, VC2 = 2.5 (bottom right).
Contour levels below 2000 are hidden to accentuate large coherent

xiii
structures. Upper wall solution has been clipped to emphasize shear layer
structures. .................................................................................................. 103
Figure 54. TKE spectrum versus wavenumber for VC2 = 1.6 solution. ........ 106
Figure 55. TKE spectrum versus wavenumber for VC2 = 2.5 solution. ........ 107
Figure 56. Coarsened backward-facing step mesh. 127 x 35 x 20 ................... 108
Figure 57. Mean streamwise velocity contours for coarse mesh solutions;
baseline laminar solution (top), VC2 = 2.0 (middle), VC2 = 2.5 (bottom).
................................................................................................................... 109
Figure 58. Mean streamwise velocity profiles from the coarse mesh comparing
experimental data, the coarse laminar solution, the VC2 = 2.0 solution,
and the VC2 = 2.5 solution. ................................................................... 111
Figure 59. Streamwise RMS velocity profiles from the coarse mesh comparing
experimental data, the coarse laminar solution, the VC2 = 2.0 solution,
and the VC2 = 2.5 solution. ................................................................... 112
Figure 60. Instantaneous center plane slices of vorticity magnitude on the coarse
mesh, from VISUALIZATION 11 with the baseline laminar solution (top),
VISUALIZATION 13 with VC2 = 2.0 (middle), and VISUALIZATION 15
with VC2 = 2.5 (bottom). ........................................................................ 113
Figure 61. Instantaneous volumetric-rendered images of vorticity magnitude on
the coarse mesh, from VISUALIZATION 12, laminar solution (top left),
VISUALIZATION 14, with VC2 = 2.0 (top right), and VISUALIZATION 16,
with VC2 = 2.5 (bottom). Contour levels below 2000 are hidden to
accentuate large coherent structures. Upper wall solution is clipped as in
Figure 53. ................................................................................................... 114
Figure 62. TKE spectrum versus wavenumber for coarse mesh VC2 = 2.0
solution. ..................................................................................................... 116
Figure 63. TKE spectrum versus wavenumber for coarse mesh VC2 = 2.5
solution. ..................................................................................................... 117
Figure 64. 5.5 case study domain with outer boundary dimensions marked. .... 119
Figure 65. Surface representation of the immersed F1 car. .............................. 120
Figure 66. Cartesian mesh used for 5.5 case study containing 3,400,000 cells;
zoomed images show the spanwise refinement through the car (top left) and
streamwise refinement along the centerline (top right). The full mesh
(centerline slice) is pictured in the bottom image. ...................................... 122
Figure 67. Reference geometry for high resolution STAR-CCM+ study including
upstream contraction. ................................................................................ 123
Figure 68. Instantaneous snapshot of vorticity magnitude from VC2 solution at
3.0 s visualized with ray-tracing technique. ................................................ 125

xiv
Figure 69. Front perspective view of vorticity magnitude iso-contour at a
magnitude of 150 for time-accurate VC solution. ....................................... 126
Figure 70. Front perspective view of vorticity magnitude iso-contour at a
magnitude of 150 for steady-state high resolution RANS solution. ............ 128
Figure 71. Side view of vorticity magnitude iso-contour at a magnitude of 150 for
the time-accurate VC solution. ................................................................... 129
Figure 72. Side view of vorticity magnitude iso-contour at a magnitude of 150 for
the steady-state high resolution RANS solution. ........................................ 129
Figure 73. Comparison between vorticity magnitude iso-contours from the VC
solution (top) and the high resolution RANS solution (bottom) for the
turbulent wake region with mesh refinement visible. .................................. 130
Figure 74. Streamline visualization based on time-averaged velocity for time-
accurate VC simulation, colored by velocity magnitude. ............................ 131
Figure 75. Streamline visualization of steady-state RANS solution, colored by
velocity magnitude. .................................................................................... 132

Figure A-1. Iso-volume of mean streamwise velocity U < 0.0 for VC1 case, 5.4
case study. ................................................................................................. 149
case study with = 1.6. ............................................................................ 149
case study with = 2.0. ............................................................................ 150
case study with = 2.5. ............................................................................ 150
Figure A-5. TKE spectrum versus wavenumber for VC1 solution, 5.4 case study.
................................................................................................................... 151
Figure A-6. TKE spectrum versus wavenumber for VC2 = 2.0 solution, 5.4
case study. ................................................................................................. 152
Figure A-7. Mesh from high resolution RANS study for 5.5 case study containing
30 million polyhedral cells; zoomed images show the local refinement around
the car (top left) and streamwise refinement along the centerline (top right).
The section of the mesh (centerline slice) corresponding to the VC2
simulation domain dimensions is pictured in the bottom image. ................ 153

xv
LIST OF ATTACHMENTS

VISUALIZATION 1: 5.4 fine no-VC Vorticity magnitude ........................... Vis1.mpg
VISUALIZATION 2: 5.4 fine no-VC Volumetric vorticity ........................... Vis2.mpg
VISUALIZATION 3: 5.4 fine VC1 Vorticity magnitude .............................. Vis3.mpg
VISUALIZATION 4: 5.4 fine VC1 Volumetric vorticity .............................. Vis4.mpg
VISUALIZATION 5: 5.4 fine VC2-1.6 Vorticity magnitude ........................ Vis5.mpg
VISUALIZATION 6: 5.4 fine VC2-1.6 Volumetric vorticity ........................ Vis6.mpg
VISUALIZATION 8: 5.4 fine VC2-2.0 Volumetric vorticity ........................ Vis8.mpg
VISUALIZATION 10: 5.4 fine VC2-2.5 Volumetric vorticity .................... Vis10.mpg
VISUALIZATION 11: 5.4 coarse no-VC Vorticity magnitude .................. Vis11.mpg
VISUALIZATION 12: 5.4 coarse no-VC Volumetric vorticity .................. Vis12.mpg
VISUALIZATION 13: 5.4 coarse VC2-2.0 Vorticity magnitude ............... Vis13.mpg
VISUALIZATION 14: 5.4 coarse VC2-2.0 Volumetric vorticity ............... Vis14.mpg
VISUALIZATION 15: 5.4 coarse VC2-2.5 Vorticity magnitude ............... Vis15.mpg
VISUALIZATION 16: 5.4 coarse VC2-2.5 Volumetric vorticity ............... Vis16.mpg
VISUALIZATION 17: 5.5 Formula One car ray-traced vorticity .............. Vis17.mpg

1
CHAPTER 1
INTRODUCTION

Essentially, all models are wrong, but some are useful.
George E. P. Box, Empirical Model-Building and Response Surfaces

Classically, fluid dynamics has been researched through experimentation
and observation of fluid motion in physical environments. The development of the
computer has led to the realization of computational fluid dynamics (CFD), the
simulation of fluid motion using numerical methods in virtual environments. The
motion of fluids is described by systems of partial differential equations (PDEs);
however, these equations do not have exact analytical solutions in their general
form. An approximate solution for a system may be obtained by discretizing the
differential equations over a finite array of discrete locations (grid points)
arranged within the domain of interest. To generate solutions over the discretized
domain requires the use of mathematical models to approximate the physical
processes involved as well as the overall behavior of the fluid. As succinctly
stated by George Box, all models intrinsically contain some amount of error due
to the approximations. Engineering judgment must be used to determine which
models are the most appropriate for a given problem in terms of accuracy and
computing speed. Therefore, the practical use of CFD as an engineering tool
involves finding the balance between solution accuracy and computational cost.
As noted by Ferziger and Peric [1], approximate solutions obtained by
numerical methods inherently contain both discretization and modeling errors.
Discretization errors are defined as the difference between the exact solution of
the system of PDEs and the exact solution of the set of algebraic equations
obtained by discretizing the PDEs over a domain. Improving the discretization
accuracy of a CFD simulation typically requires the use of higher order methods,
which can be numerically unstable, or an increase in the number of grid points

2
within the solution domain. As the grid becomes refined, the discretization error
decreases and the truncation error of the approximate solution approaches a
minimum limited by the order of accuracy of the solution method. Increasing grid
points, however, adds considerable computational cost through the amount of
time required to solve the discretized equations and typically requires large
computational resources. For a complex engineering problem this increase in
cost may be unfeasible due to calendar and budget constraints.
Modeling errors are defined as the difference between actual flow
behavior and the solution of the mathematical model used to describe the flow
behavior. It is important to realize that even with minimum discretization error the
exact solution of the mathematical model may be physically incorrect if actual
flow behavior deviates from the modeling assumptions. Traditional CFD has
applied models with empirical constants or parameters which are calibrated such
that the resulting solution becomes agreeable to experimental data. However,
accurate results through calibration are obtainable only for similar types of flow
problems, typically with simple geometries. As the physical processes become
more complicated, accurately approximating the real flow behavior with a
calibrated model becomes more difficult. Typically conventional models have
difficulty resolving thin convecting flow structures such as vortices and vortex
sheets, which generally dissipate due to the numerical diffusion introduced by
modeling and discretization errors.
As an alternative to traditional modeling approaches, Steinhoff [2] has
developed a numerical technique called Vorticity Confinement (VC). VC is an
Eulerian fixed-grid methodology which exhibits similar characteristics to
Lagrangian vortex methods in treating vortices and other thin vortical flow
features. VC can work to confine thin regions of high vorticity to 2-3 grid cells and
prevent the spreading of vortical structures due to numerical diffusion as they
convect through the flow field. In this manner, the effects of the smaller flow
structures on the bulk fluid are captured without modeling the details of the
structures directly. The basic technique involves adding a forcing term to the

3
momentum equations which acts within the limit of vortical regions. The behavior
of VC is governed by two parameters, and , representing the balance between
isotropic diffusion and isotropic compression, respectively, applied to vortical
regions. The application of VC to interacting vortex rings has demonstrated the
ability of the method to accurately model merging and interaction of vortices over
coarse grids [2,3]. Additional studies have shown considerable promise for the
application of VC to a variety of flow conditions, and will be discussed further in
detail in Chapter 2 [4,5,6,7]. The VC methodology is intended to provide a
simple, efficient model which can be applied to many types of flow features such
as boundary layers, free shear layers, and separated flows with minimal
computational overhead.
1.1 Objectives
The inherent characteristic of potentially higher accuracy for lower
computational cost suggests that Vorticity Confinement is an ideal model for fast
prototyping and simulation during the engineering design process. However, to
date the method has not been widely utilized in mainstream CFD simulation
within the engineering community. This is possibly due to a lack of clarity
regarding the proper range of values for the confinement parameters. Previous
studies have suggested that the effects of VC are subjective and that the
confinement parameters may be problem dependent [8,9,10]. More research is
needed to evaluate the dependencies and behavior of the method and determine
consistent values for the confinement coefficients which can apply to a wider
range of flow conditions.
To perform simulations necessary for such an investigation, the primary
focus of this thesis is the development and validation of a fully parallelized
Vorticity Confinement solver using the OpenFOAM (Open Field Operation And
Manipulation) computational framework. OpenFOAM is an open-source
collection of C++ libraries, classes, and methods which together form a fully
functional CFD package based on finite volume discretization. OpenFOAM has

4
seen extensive use in the academic community, and in recent years many
commercial companies have begun using OpenFOAM for CFD simulation
supporting a variety of industries [11]. The open-source nature of the package
encourages users to freely share source code with the extended OpenFOAM
community, thereby benefitting all users equally. The presence of a VC solver
within OpenFOAM would potentially create interest in more applications of VC to
different flow problems and expose a wider portion of the engineering community
to the benefits of the methodology. Furthermore, tapping into the OpenFOAM
user community would provide a resource for the calibration of the VC model
parameters for a wide range of engineering problems.
After developing the flow solver, the complementary focus of this thesis is
the exploration of the effects and behavior of Vorticity Confinement through
simulation of both simple benchmark cases as well as more complex flow
configurations. It is intended through this course of study to determine acceptable
ranges of the confinement parameters for realistic flow simulation. Ultimately,
with sufficient calibration and understanding of the confinement parameters it is
hoped that the developed flow solver can be utilized as a production CFD code.
1.2 General Approach
The content of this thesis is separated into chapters pertaining to specific
topics. Chapter 2 focuses on the development and background of the Vorticity
Confinement method. The VC1 and VC2 formulations are discussed and various
extensions of the method are treated in detail, including dimensional and
dynamic confinement, in order to give the reader a full overview of all
developments using the method. Chapter 3 gives an introduction to the open-
source CFD simulation package OpenFOAM. This chapter serves primarily as a
record for various details learned during the course of study regarding the
workings of OpenFOAM. An overview of the general structure of the included
libraries is presented along with a description of programming using OpenFOAM.
A typical case file structure is outlined in order to provide the reader a level of

5
familiarity with the package. Chapter 4 describes the finite volume discretization
of the conservation equations used within the developed flow solver and provides
an overview of the pressure-correction methodology used in OpenFOAM to treat
time-accurate and steady-state simulation. The implementation details of the VC
source term and a parallel harmonic mean subroutine necessary for the VC2
formulation are described along with the immersed boundary formulation
implemented for use with VC. Chapter 5 highlights verification of the flow solver
with the presentation of results from five simulations conducted with the solver. A
flat plate boundary layer is simulated in 5.1 to show the effect of VC on boundary
layer flows, while in 5.2 an immersed slanted plate simulation demonstrates the
viability of the implemented immersed boundary method. The effect of VC on an
advecting compact vortex is shown in 5.3. Validation of the flow solver is
performed in 5.4 with the comparison of a backward-facing step simulation to
experimental data. The case study in 5.5 demonstrates the viability of the flow
solver to perform fast three-dimensional complex flow simulation by modeling a
Formula One racecar within a wind tunnel domain using the immersed boundary
method. Chapter 6 discusses the subsequent findings and conclusions from this
work, as well as future plans of study.

6
CHAPTER 2
VORTICITY CONFINEMENT

As stated in Chapter 1, Vorticity Confinement (VC) is a numerical
technique conceived by Steinhoff [2], which has been developed to model thin
vortical flow structures and convect them over long distances while minimizing
dissipation. VC emerged from earlier work in the late 1980s with vortex capturing
methods such as the Vorticity Embedding technique, which was developed and
validated for hovering rotor studies in helicopter research [12,13,14]. Similar to
vortex capturing methods, VC does not attempt to resolve the internal features of
a vortical structure; the Eulerian VC model confines the internal structure to 2-3
grid cells and prevents dissipation while preserving the general effect of the
overall structure on the surrounding flow field. This can have the effect of
reducing the computational requirement for complex flow simulations by enabling
the resolution of vortical flow features on a coarser grid than required by other
modeling techniques. VC acts through a correctional forcing term within the limit
of the vortical regions and has no direct effect in the far field non-vortical solution.
Furthermore, the method does not require any special discretization scheme, and
can be used with both structured and unstructured grids, making it possible to
integrate VC into established flow solvers.
Steinhoff and co-workers have developed two formulations of the Vorticity
Confinement term which they name VC1 and VC2, respectively [15]. The VC1
formulation, which uses a first-order derivative in the confinement term, is treated
in Section 2.1. Surface Confinement, a special formulation of VC1 developed for
boundary layer flows, is discussed in Section 2.1.1. Section 2.2 introduces the
VC2 formulation, an alternate form of the confinement term which uses a second-
order derivative and explicitly conserves momentum [16]. A formulation of
Vorticity Confinement has also been developed for compressible flows and is
described in Section 2.3; while the compressible flow regime is out of the scope

7
of this thesis, the formulation describes a method to introduce the confinement
term to the integral form of the momentum equations, which is of central
importance with respect to the finite volume formulation used in OpenFOAM.
Finally, Section 2.4 details research of alternate formulations achieving a
dynamic confinement strength parameter , as opposed to the constant
parameter used by Steinhoff in the original formulation.
2.1 VC1 Formulation
Much of the available literature concerns studies using the VC1
formulation. Steinhoff and Underhill [3] showed excellent agreement to
experimental data for the collision of two coplanar vortex rings in a three-
dimensional time-accurate calculation. With the addition of the confinement term,
the rings were shown to automatically merge while preserving local vorticity
magnitude. Vorticity Confinement has been applied to a variety of incompressible
complex flow configurations using body-fitted grids as well as immersed
boundaries on uniform grids. Steinhoff, Wenren and Wang [4] presented
encouraging results for a pitching airfoil in dynamic stall, a Boeing 747 wing, and
an Apache helicopter in forward flight. Simulations involving a helicopter fuselage
[17], wake flows behind a three-dimensional cylinder and representative aircraft
carrier [5], and flow over a 6:1 ellipsoid [18] exhibited good correlation to data
(when available for comparison) and improved vortical flow predictions with the
inclusion of the confinement term. A recent study implementing VC1 into a
proprietary OpenFOAM solver showed a reasonable correlation to data for flow
over an Ahmed body vehicle shape with various slant configurations,
demonstrating the applicability of VC for blunt body flow separation [10].
Recently, VC1 has been implemented in a smoothed particle hydrodynamics
(SPH) model for bubble prediction [19] and in a two level simulation (TLS) flow-
scale separation method for water turbulence simulation [20], both of which
demonstrate the versatility of Vorticity Confinement to integrate with various
solution routines.

8
The basic VC methodology involves the addition of a discrete term to the
discretized momentum conservation equations. The vector form of the
conservation equations for unsteady, incompressible, isothermal flow, including
the vorticity confinement term, is shown in Equations (2.1) and (2.2):

(2.1)

( ) ( ) () (2.2)

where U is the velocity vector, p is pressure, is density, and is kinematic
viscosity (equivalent to /). The confinement strength coefficient is represented
by . The parameter represents both physical viscosity and explicit numerical
viscosity necessary for the VC formulation. Dissipation multiplied by and anti-
dissipation multiplied by automatically balance over a range of values. The
confinement term takes different forms depending on the confinement
formulation. For the VC1 formulation, the corresponding term is shown in
Equation (2.3):

(2.3)

where is a unit vector and is vorticity, calculated as the curl of the velocity
field. For confinement in a vortical flow field away from a solid boundary, takes
the form in Equation (2.4):

||
||||
(2.4)

This is a unit vector which points towards the maximum gradient of
vorticity. Effectively, the confinement term serves to convect vorticity back along
the unit normal towards regions of high vorticity. This action energizes the
vortical regions of the flow field, such as vortices and shear layers, and acts as a

9
counter to numerical dissipation inherent within the solution. It has been shown
by Steinhoff [2] that with VC both mass and vorticity are globally conserved. It
should be noted that the VC1 formulation does not conserve momentum exactly
due to the self-induced velocity of vortical regions; however, this error is minimal
for most problems and becomes significant only when long-term vortex trajectory
accuracy is a concern [7]. The later development of the VC2 formulation,
discussed in Section 2.2, provides an alternative that explicitly conserves
momentum.
2.1.1 Surface Confinement
As previously noted, a special formulation of VC1 was developed for
boundary layer regions [4,17]. This form explicitly convects vorticity back towards
the boundary surface, defined by a function F. If an immersed boundary method
is used, then F is a smooth level set function which is defined as 0 at the
boundary surface, positive outside the boundary in the flow field, and negative
inside the boundary (interior of the immersed body). If body fitted grids are used
then F can represent the distance to the wall. Studies of the application of
Surface Confinement to missile aerodynamics showed improvement over simple
Euler solutions, and some calibrations of the confinement strength parameter
have been carried out for this specific application [21,22]. With Surface
Confinement the following form of is used in cell layers near the boundary:

||
||||
(2.5)

In theory, the form of in (2.4) should work in the boundary layer as well,
since it acts to convect vorticity towards the highest local value which should be
at the no-slip wall. However, in practice the form in (2.4) can become unstable
close to the wall as the confinement strength coefficient is increased. The form of
in (2.5) explicitly directs vorticity towards the wall and helps stabilize VC1 in the

10
boundary layer. The VC2 formulation, discussed in the following section, does
not require any such explicit treatment in the boundary layer.
2.2 VC2 Formulation
The elements comprising the VC2 formulation were first introduced as a
method to treat thin vortical flow features as solitary waves, such that the
internal structures remain thin over a minimal region 2-4 grid cells wide,
irrespective of time or distance traveled [23]. Steinhoff et al. [16] formalized the
method for multidimensional calculations as a new Vorticity Confinement
formulation (VC2) which explicitly conserves momentum for vortical regions.
Results for simple three-dimensional blunt bodies (cylinder, sphere, and a half-
sphere shell) indicated better vortical flow prediction using the new VC
formulation which simulated higher Reynolds number effects [6]. Additional
studies detailed by Steinhoff and Lynn [7] and Steinhoff et al. [18] concerning
wing tip vortices and wake profiles behind various simple bluff bodies showed
good agreement with experimental data as well as VC1 simulation results (where
available). Lynn [24] developed turbulence models based on the VC2 framework
which automatically determine an appropriate confinement strength based on
conservation of kinetic energy, similar to dynamic sub-grid scale LES models,
and validated the models against experimental and numerical solutions for the
Taylor-Green vortex, a flat plate boundary layer, and a backward-facing step. A
higher-order extension of the VC2 formulation has also been developed recently
and has been verified against analytical pulse solutions [25].
As stated previously, the VC2 formulation has been developed to explicitly
conserve momentum which is a shortcoming of the VC1 formulation. The
formulation introduced by Steinhoff involves a similar additional confinement term
as VC1 in Equation (2.2) with a new definition:

(2.6)

11
The term in the cross product is given for the discretized momentum equations
as follows:

()
(2.7)

() [
(

(2.8)

|| (2.9)

The
term in (2.7) can be interpreted as the normalized vorticity vector

multiplied by the harmonic mean in (2.8). The value in (2.9) is a small constant
added to prevent division by zero in the
term in (2.7). The harmonic mean is

calculated over a localized stencil of N cells; for a uniform hexahedral mesh with
face-based cell connectivity, this involves the center cell and all 4 (2D) or 6 (3D)
neighboring cells. Depending on the underlying structure of the computational
code (i.e., node-based connectivity) additional neighbors are possible in the
harmonic mean calculation.
2.3 Compressible Vorticity Confinement
Traditional compressible flow solvers solve the conservation equations as
a coupled system, rather than using the split-velocity methods popular for
enforcing conservation in incompressible flow. As a result, the velocity correction
used in incompressible VC must be introduced as a source term. A consistent
methodology for compressible Vorticity Confinement was first suggested by Hu
[26]. Hus research suggests that the confinement term may be interpreted as a
body force, which can then be extended directly to the integral momentum
equations and the rate of work done by the body force may be taken into account
in the energy conservation equation. Previously, VC had been implemented in
finite difference schemes; the integral formulation allows extension to finite-

12
volume schemes which form the basis of many commercial CFD codes. Hu and
Grossman [27] displayed overall good quantitative and qualitative comparison to
experimental data and numerical LES simulations for massive separation from
bluff bodies using their compressible Vorticity Confinement formulation and
uniform coarse grids. For uniform refined grids and an equivalent confinement
parameter , however, the comparison to experimental data indicated some
dependency of on the grid density. Jafarian and Fard [9] further extended the
compressible Vorticity Confinement formulation suggested by Hu with variable
confinement parameters as functions of the spectral radii of the Jacobian
matrices and the Jacobian matrices themselves. This extension, however, still
required tuning the confinement parameter, although the resulting range was
smaller than in the original compressible formulation.
In contrast to the compressible Vorticity Confinement suggested by Hu,
Costes [28] showed substantial evidence that in fact the correct application of the
confinement term to compressible flow does not require the additional work term
in the energy conservation equation. Costes noted that, for the advection of a
two-dimensional vortex, including the additional term in the energy equation
resulted in a diffusion of density towards the exterior portion of the vortex.
Removing the source term from the energy equation recovered the correct
density prediction. This is not entirely surprising, as Hu [26] noted in the original
research that this additional term had detrimental effects on total temperature.
Costes surmised that associating physical quantities such as work from the body
force is unnecessary because Vorticity Confinement is a numerical treatment
without a physical analog.
While compressible flow is outside the scope of this thesis, the principles
developed concerning the application of Vorticity Confinement to the integral
momentum equations, and thereby finite-volume methods, will be followed. The
finite-volume formulation detailed in Chapter 4 adheres to the suggestion of
Costes [28] with the addition of the confinement term to the momentum
equations alone.

13
2.4 Dimensional Analysis of
Previous studies [8,9] have indicated that inconsistent values of can
result with the VC1 formulation when varying the mesh density or flow conditions.
In the VC1 formulation, the confinement parameter introduced in Equation (2.2)
requires the dimension of velocity to ensure consistent dimensions with the other
conservation equation terms. Towards the derivation of a consistent parameter,
Fedkiw et al. [29] introduced a dependency on mesh size, h, such that the
confinement strength scales with overall mesh size and varying mesh refinement.
Lhner et al. [30] performed dimensional analysis on the confinement parameter
and determined three potential scale factors which ensure a dimensionless
confinement parameter for VC1:

|| (2.10)

|| (2.11)

|||| (2.12)

Lhner et al. determined that the scaling in Equation (2.12) is desirable
because it cancels the denominator of the unit normal vector in Equation (2.4),
reducing the number of mathematical operations necessary for the confinement
term. They suggested a refined length scale h based on the characteristic length
in the direction of the vorticity gradient, as well as an explicit cutoff threshold
based on a local vorticity-based Reynolds number to remove the effect of
Vorticity Confinement from highly refined boundary layer regions. Robinson [31]
further extended the work of Lhner by factoring out vorticity magnitude from the
confinement term, and using the dimensional scaling form in Equation (2.10).
Robinson noted that the product of velocity and vorticity magnitude has identical
dimensions to helicity, and suggested an alternate VC1 confinement term with
the quantity of helicity as follows:

14

| | [
||
||||
||
] (2.13)

For Surface Confinement, rather than using the cutoff threshold suggested
by Lhner, Robinson used the scaling term in Equation (2.12) to take into
account mesh size reduction in the boundary layer region. Using this combination
of confinement terms Robinson showed improvement for missile force and
moment predictions using the formulation. The Robinson formulation has a key
weakness, however, in that helicity is zero for two-dimensional simulations or for
vortical structures normal to the freestream flow direction (tornado-like
structures). This deficiency is not present in the Lhner scaling formulation.
Furthermore, scaling the confinement parameter with helicity alone does not
provide a dependency on mesh size which is necessary when using a non-
uniform mesh. Butsuntorn and Jameson [32,33] extended Robinsons helicity
formulation to compressible flow rotorcraft calculations by including a non-
dimensional scale factor based on mesh size:

| | [
)

] [
||
||||
||
] (2.14)

The formulation suggested by Butsuntorn and Jameson showed modest
improvement to lift and drag coefficient predictions for rotorcraft flow [32]. They
noted that the confinement parameter was dependent on the numerical scheme
used in each calculation; as a result, because different numerical schemes have
varying discretization error the confinement parameter value may vary by up to
one order of magnitude between simulations. They suggested that the
confinement parameter should be scaled based on numerical diffusion and
discretization errors to improve the consistency of calibrated confinement
strength values.

15
2.4.1 Dynamic Vorticity Confinement
As suggested by Butsuntorn and Jameson [32], for a consistent
formulation to counter numerical error the confinement parameter should be
based on numerical diffusion and discretization errors present for the given
discretization scheme. Costes and Kowani [34] developed for VC1 a dynamic
confinement parameter , constant over each grid cell, which was based on the
assumption that the confinement term must cancel the truncation error of the
momentum equations. This definition also resulted in the correct dimensionality
of the confinement parameter. However, due to compressibility effects (the study
was in the compressible flow regime) and the singularity of their formulation at
the vortex center, they observed anomalies in the vorticity and density contours
near the center of the vortex. Costes [28] reported improved results for a similar
study using the VC2 formulation and a modified dynamic confinement formulation
which resulted in the correct dimensionality for the VC2 confinement parameter
and a constant value for the confinement strength averaged over the entire
vortex.
Hahn and Iaccarino [35] detailed a similar approach towards creating a
dynamic confinement parameter. They reasoned that the primary purpose of the
VC term in the conservation equations is the cancellation of numerical diffusion
and discretization errors; to that effect, they applied principles from dynamic LES
modeling to determine a dynamic confinement strength and estimated numerical
viscosity to be equivalent to the difference between central and upwind
discretization schemes for the convection term. Results for the collision of two
vortex rings and a counter-rotating vortex pair rebounding from a wall showed
excellent agreement to analytical and experimental data using this formulation.

16
CHAPTER 3
INTRODUCTION TO OPENFOAM

OpenFOAM (Field Operation And Manipulation) is an open-source
collection of C++ class libraries developed for simulating continuum mechanics
[36]. The CFD package originates from work begun in the 1990s at Imperial
College in London under Prof. David Gosman and Dr. Raad Issa [37]. The
principal developers, Henry Weller and Hrvoje Jasak, began marketing FOAM
and in 2005 moved to two separate paths of development utilizing the open-
source business model [37]. Weller maintains the official version of OpenFOAM
which has closed-source development and periodically releases open-source
version updates [38]. Jasak maintains a fork of OpenFOAM called the Extend-
Project (also known as OpenFOAM-extend) which includes additional modeling
capability not found in the official version and is centered on community-driven
development [39]. The solver developed in this thesis utilizes libraries from
OpenFOAM-extend version 1.6.
One of the difficulties experienced during the development of the current
work was gathering the necessary knowledge to operate pre-existing
OpenFOAM solvers as well as develop a new Vorticity Confinement solver for
this work. OpenFOAM, like many open-source projects, is sparsely documented
and the included manuals provide an incomplete description of components
within the CFD package. Therefore, this chapter serves as a record of the
information learned from various sources during the development of this work
and provides a brief introduction to OpenFOAM with the intention of explaining
specific fundamentals to the uninformed reader. Key aspects of the C++
programming language which are used throughout the libraries of OpenFOAM
are presented in Section 3.1. The source code is an excellent example of proper
C++ programming practices and can serve as its own documentation when
included functionality is not described by the manuals. Key programming

17
concepts used in OpenFOAM are presented in Section 3.2. The OpenFOAM
library structure and mesh structure are discussed in Section 3.3. Case file
structure aspects which are pertinent to setting up, solving, and post-processing
a simulation are presented in Section 3.4. Finally, descriptions of common
utilities included in OpenFOAM are given in Section 3.5.
3.1 Introduction to Object-Oriented Programming
An explanation of the basic concepts of object-oriented languages is
necessary to highlight the advantages of the C++ programming language.
Object-oriented programming (OOP) typically makes use of the following
concepts [40]:
Classes and Objects a class defines a data structure and all of the
functions which operate on the data structure. An object is an instance
of a class (instantiated at run-time). Objects share the same functions
with other objects of the same class, but each object contains a copy
of the original class data structure.
Member functions these are functions which are declared inside the
class definition and are carried as part of the object. Two special
member functions exist for each class: the constructor, which is called
on initialization of the object, and the destructor, called on deletion of
the object.
Private vs. Public members both member functions and data within
a class require access permissions. A public member can be read or
written by any function, for data, or called by any object, for functions.
Private members can only be used by objects of the same defining
class.
These concepts allow efficient organization and design of programs and
are essential when dealing with very complex systems. Embodying the OOP
paradigm, the C++ programming language was developed in the early 1980s by
Bjarne Stroustrup with the aim of combining the flexibility and efficiency of C

18
systems programming with OOP paradigms [41]. Support for the following
paradigms is a cornerstone of the C++ programming language:
Data abstraction: focuses on user-defined data types, in contrast to
built-in types. Abstraction requires that a full set of operations be
provided for each type of data [42]. Central concepts to this style are
operator overloading and polymorphism. Operator overloading allows
the user to define specific actions when using built-in operators with
user-defined data types. Polymorphism is the ability to define
an abstract base interface which applies to multiple objects at run-time.
Object-oriented programming: introduces the concept of inheritance
with class hierarchy. A base class defines generic properties of a class
of objects and is specialized by derived classes which inherit the base
class properties. Classes contain both data and functions, providing a
method of encapsulation in which the classes are self-contained
objects. This concept takes advantage of commonality in systems by
using classes as building blocks to construct more complex objects.
Generic programming: allows for algorithms to be parameterized for
working with a variety of types and data structures, such that the
algorithms are independent of the representative details [42]. Generic
programming introduces the concepts of templates and container
classes. Templates are classes or functions which are written with
generic types, allowing them to work with multiple data types without
being rewritten for each type. Containers are classes that can hold a
collection of other objects.
While C++ was designed to support these programming paradigms in
addition to traditional C systems programming techniques, it does not force any
one particular style on all users [42]. The language simply makes it convenient to
use any standard programming paradigm which fits the task at hand. This
freedom is part of what makes C++ such a powerful scientific programming
language.

19
3.2 Programming in OpenFOAM
By taking advantage of the aforementioned programming concepts, the
developers of the OpenFOAM C++ library have taken a unique approach to
solving PDEs by mimicking standard vector and tensor notation with high-level
code, thus allowing complicated mathematical and physical models to be
represented by high-level mathematical expressions [36]. OpenFOAM includes
such C++ concepts as the use of polymorphism to represent subsets of
turbulence models and boundary conditions with similar code interfaces,
inheritance with a base class for mesh data and derived classes for access, and
templates allowing code re-use for operations on different data types [36,43].
As an example of an OpenFOAM representation of a PDE, the momentum
equation for unsteady isothermal flow is given as an example in Equation (3.1):

(3.1)

The OpenFOAM code snippet in Figure 1, identical to Equation (3.1), is
valid C++ syntax using the OpenFOAM high-level classes [44]. Operator
overloading allows tensor algebra to be defined using basic mathematical
operators for easily understandable equation syntax. Even though some of the
syntax may not be immediately recognizable to one unfamiliar with OpenFOAM,
the code clearly describes a time derivative term, a divergence term (convection),
a laplacian term (diffusion), and a gradient term, and therefore provides an exact
description of the discretized PDE. The terms in the equations are constructed to
behave like their mathematical counterparts and hide the numerical details of the
implementation by encapsulation [36]. Solution variables such as velocity and
pressure are represented by tensor field data types which can describe scalars,
vectors, and tensors. All physical variables are denoted by tensor fields sharing
meaningful names: rho for density, mu for viscosity, U for velocity, and p for
pressure. This design allows the user to focus on the correct description of the

20
PDE to be solved without worrying about discretization details which remain
encapsulated within the high-level code. Furthermore, when dealing with
unfamiliar models the code syntax can serve as its own documentation.
3.3 OpenFOAM Data Structure
The OpenFOAM lower-level libraries outline generic interfaces for data
input and output to ensure consistency throughout the simulation environment
and provide re-usable components as building blocks for higher-level libraries.
For example, the lower-level libraries define discretization and parallelization
details which are then implicitly used in higher-level code. Standalone
executables are generated for utilities and solvers; generally, standalone solvers
are written to manage one specific physics regime, such as buoyancy simulation
or combustion. Multi-physics simulation requires combining existing solvers to
create new customized solvers. This component-based approach gives the user
a great amount of flexibility which may not be available in commercial CFD
codes, and the availability of all source code eliminates the black box factor in
commercial codes in which the details of model implementation are hidden.
To discretize the problem domain, OpenFOAM uses an internal
unstructured mesh format which supports tetrahedral, hexahedral, and
polyhedral cell types. All variables are defined within the same control volumes,
an approach which is typically called a collocated or non-staggered grid

Figure 1. OpenFOAM C++ function call identical to Equation (3.1).

21
arrangement [1]. All meshes are explicitly three-dimensional; one- or two-
dimensional calculations are performed by using special boundary conditions on
any plane(s) normal to the coordinate direction(s) of interest which switch off the
solution algorithm in the unnecessary direction(s). OpenFOAM requires the
following mesh quality criteria to be met, as stated in the official User Guide [44],
for a mesh to be computationally valid:
Contiguous cells must cover the full computational domain and not
overlap one another.
Convex every cell must be convex such that the cell center must be
contained within the cell.
Closed each cell should be closed, such that the sum of all face area
vectors oriented outwards for each cell should equal zero, and each
edge is used by exactly two faces of the given cell.
Orthogonality for each internal face, the angle between the face
area vector and the center-to-center vector must be less than 90.
Figure 2 and Figure 3 detail a polyhedral cell illustration and its
corresponding OpenFOAM mesh representation. As shown in Figure 3, the
polyhedral mesh in OpenFOAM is represented with four linked lists: a list of point
coordinates, a list of cell faces with references to the point coordinates, a list of
owner cells for each face, and a list of neighbor cells for each face [45]. This well-
defined approach is optimal for interfacing with the unstructured mesh during the
solution process.
3.4 Standard Case Structure
An OpenFOAM simulation is defined by a specific file structure which
holds the setup information, mesh definition, and solution files. The standard file
structure for a serial (single processor) case is shown in Figure 4. Each
simulation directory holds three sub-directories: the system, constant, and time
directories (named using the physical time step or iteration step identifier).

22

Figure 2. Two cell polyhedral mesh [45].

Figure 3. OpenFOAM linked list mesh representation of two cell mesh in Figure 2 [45].

23

Figure 4. OpenFOAM case directory structure.

24
3.4.1 system Directory
This directory contains discretization and solution procedure information.
General solver and utility settings in OpenFOAM are controlled by dictionary files
which have the -Dict suffix. These files are read dynamically during solver
execution and allow parameters to be varied during runtime. The following files
are located within the system directory:
controlDict this dictionary contains run control parameters, such as
start/end time, time step controls, and data output settings. Additional
data output can be defined here using pseudo-code user functions
described in Section 3.5.5.
fvSchemes this file defines the discretization schemes used for each
algebraic tensor operation and each high-level mathematical
expression.
fvSolution the iterative equation solver settings, tolerances, and
specific algorithm control parameters are defined within this file.
decomposeParDict this file defines domain decomposition
parameters for use with the decomposePar tool for parallel simulation.
3.4.2 constant Directory
The constant directory holds the mesh description and any files with the
-Properties suffix describing required physical properties, such as viscosity.
Additionally, turbulence model parameter files are located here if necessary for
the simulation. The polyMesh directory contains the following mesh
representation files:
points the points file is an ordered list of vectors describing the
three-dimensional location of each point of the mesh in sequential
order. Each row contains the global x, y, and z coordinates of the point
in dimensions of meters.
faces the faces file contains the faces defined by the vertices in the
points file. Each row of the list corresponds to a face in sequential

25
order and is comprised of a counter indicating the number of vertices
making up the face, followed by a vector with the corresponding vertex
labels. Because OpenFOAM supports polyhedral meshes, faces can
have more than three (tetrahedral mesh) or four (hexahedral mesh)
points.
owner each face in OpenFOAM is associated with an owner cell
and a neighbor cell. The owner file is a sequential list of cell numbers
which directly corresponds to the list in the faces file.
neighbour the neighbour file is a sequential list of cell numbers,
similar to the owner file, except the listed cell numbers correspond to
the neighbor cells for each face. Boundary faces are given a -1 index
within this file as each boundary face belongs to an owner cell only.
boundary this file contains a list of patches (geometrical definitions
of the boundaries) for the simulation. The number of corresponding
patches is required at the beginning of the list. Each patch is given a
name and described by a dictionary containing three entries. The
geometrical definition of the patch is the type entry, and takes values
of patch (used to apply most boundary conditions), wall (denotes
boundaries for distance to wall calculations), and empty (defines
patches to be ignored for one- or two-dimensional calculations). The
faces comprising the patch are defined by a beginning face index with
the startFace entry and the number of subsequent faces in the
nFaces entry.
The polyMesh directory also contains the blockMeshDict file which is used
for input to the blockMesh mesh generation utility. This utility will be discussed in
greater detail in Section 3.5.2. The other directory in the constant directory, but
not required at simulation runtime, is the triSurface directory. Any necessary
surface definitions, such as Stereolithography (STL) files, are located here.

26
3.4.3 Time Directories
The time directories are named using the specified numerical format for
the time step in the controlDict file in time-accurate simulations, or the iteration
number in steady-state simulations. All data necessary for post-processing the
simulation and for restarting at a specified time step or iteration are contained
within the time directories. These directories are created dynamically by the
running solver as the solution is generated; if a directory already exists, the
solution files contained within the existing directory will be overwritten.
For each simulation, initial conditions are defined within the 0 directory.
Every initialized simulation field is required to have a corresponding file in the 0
directory; for example, the velocity and pressure fields are defined by U and p
files, respectively. Each file follows a specific formatting scheme. Dimensions of
each field are specified by a vector which corresponds to a list of SI units of
measure. The initial conditions of the field are specified with an internalField
entry. Boundary conditions on each patch for each field are set with a
boundaryField entry which lists the patches defined in the boundary file in the
polyMesh directory. Each boundary condition requires various inputs which are
discussed in further detail in the official User Guide [44].
Each subsequent time directory after the 0 directory contains all outputted
simulation field results in a single file for each field for post-processing or for
restarting the simulation. The outputted fields follow the same file format
described above; boundary condition information for each field is carried in a
boundaryField entry as in the initial condition files. In each field file, simulation
data for all cells are recorded following the internalField entry in a sequential
list based on cell number. This enables the same files to be used for both
simulation restart as well as post-processing results from the simulation.
3.5 OpenFOAM Utilities
An assortment of utilities which provide various levels of functionality is
included with the standard OpenFOAM package. Each utility is a separate

27
executable which is called within the case directory to perform certain operations.
Sample dictionary files are incorporated with most utilities which allow utility
settings to be specified. Pre-processing utilities cover field mapping between
meshes, setting initial conditions, and other various initialization functions.
Meshing utilities are included for mesh generation, mesh conversion, and mesh
manipulation. Post-processing utilities cover data conversion to various third-
party packages, basic field calculations, and data sampling. Some of the
common utilities are discussed in this section; a complete list of utilities is
published in the official User Guide [44].
3.5.1 Pre-processing
Most of the pre-processing utilities perform initialization of fields for
various purposes. The setFields utility allows the user to set non-uniform initial
values of fields based on bounding shapes or other parameters defined in the
setFieldsDict file. The createTurbulenceFields tool creates and initializes all
of the fields necessary for turbulence modeling. Another important pre-
processing tool is the mapFields utility, which allows fields to be mapped
between meshes sharing similar geometry, or from a coarser mesh to a finer
mesh.
3.5.2 Meshing
Basic mesh generation is provided by the blockMesh and snappyHexMesh
utilities. The blockMesh utility allows the user to generate multi-block meshes
and reads a geometry definition comprised of points, blocks, and patches from
the blockMeshDict file in the polyMesh directory as mentioned in Section 3.4.2.
Blocks are specified by listing the points comprising each block according to the
right-hand rule and defining grid spacing requirements [44]. For a typical user it is
relatively difficult to generate complex meshes using blockMesh because the tool
is based fully on text file input and has no graphical interface to visualize the

28
designed mesh. However, the tool performs adequately for simple meshes or for
an initial mesh as a basis for trim meshing with snappyHexMesh.
The snappyHexMesh utility provides an alternative method for generation
of three-dimensional body-fitted hexahedral grids. The tool is inherently three-
dimensional and requires an initial mesh describing the computational domain
and a surface representation of the complex body in STL format. From the initial
base mesh, snappyHexMesh automatically generates a mesh conforming to the
surface with optional snapping of cells to the surface and optional extrusion layer
generation. The utility also allows for parallel mesh generation with a load
balancing step during each iteration [44].
If the mesh has already been generated using third-party meshing
software, OpenFOAM provides many utilities to convert the mesh to the
OpenFOAM mesh format. Conversion utilities exist for meshes from the primary
commercial CFD packages of Fluent, STAR-CD, STAR-CCM+, and CFX.
Additional conversion is possible from meshes generated by standalone meshing
suites such as Gambit, Pointwise and IcemCFD.
After converting a mesh into OpenFOAM format, many utilities exist for
modifying, fixing, or otherwise preparing a mesh for a simulation. There are
utilities to refine, renumber, or rotate the mesh, as well as create baffles or
additional patches. OpenFOAM also allows the definition of cellZones and
faceZones, special sets of cells or faces which are automatically decomposed
for parallel processing.
3.5.3 Domain Decomposition
Parallel calculations require domain decomposition, which is the process
of splitting the computational mesh for allocation among a number of processors.
Communication between processors is accomplished through the use of a
Message Passing Interface (MPI) which is a communications protocol for high-
performance parallel computing. OpenFOAM is built with the open-source

29
OpenMPI protocol by default; however, other MPI protocols can be supported
with some minor changes in the package compilation process.
The process of splitting the domain for parallel computation is performed
in OpenFOAM by the decomposePar utility. Several mesh splitting algorithms are
available. Simple decomposition allows the user to define split regions by
equivalent sectioning of the domain based on the Cartesian directions. Metis
and Scotch decomposition are standardized partitioning algorithms based on
the minimization of the number of interface boundaries for each processor.
The decomposePar utility is run in the case directory and generates local
processor directories for every processor. Each directory contains separate time
directories which hold the field values corresponding to cells allocated to that
processor. During the solution process, each processor writes out the necessary
fields into its local directory. After the simulation is complete, executing the
reconstructPar utility in the case directory recombines all of the solution files
into common time directories in the case directory.
3.5.4 Post-processing
The standard OpenFOAM installation is packaged with a post-processing
utility called Paraview, an open-source data visualization package produced by
Kitware and based on the Visualization Toolkit (VTK) format for data rendering
and post-processing. Paraview version 3.10 includes a built-in data reader plugin
for OpenFOAM data which provides an option for working directly with
decomposed data, eliminating the reconstructPar step described in Section
3.5.3. For large cases a foamToVTK utility is available to convert OpenFOAM
data directly to the VTK format for faster post-processing if necessary. If other
post-processing tools are available OpenFOAM includes utilities to convert data
to popular third-party post-processing utilities such as EnSight, Fieldview, Tecplot
and Fluent.
In addition to visual post-processing, OpenFOAM also includes utilities for
numerical post-processing of simulation field files. Utilities such as Mach and

30
streamFunction calculate quantities from the stored velocity field in each time
directory and output additional field files for the calculated values. Tensor
operations such as divergence, magnitude, and splitting of a field into
components are available to operate on stored field files using the foamCalc
utility. Lastly, solution sampling can be conducted on stored field files at defined
survey locations in the domain using the sample and probeLocations utilities.
3.5.5 Run-time Processing
In addition to post-processing, OpenFOAM allows users to conduct run-
time processing using function objects. Function objects are pre-compiled
libraries which can be added to the controlDict file to perform various processing
tasks during run-time. The range of available functionality is extensive and
includes sampling of data at point locations and along survey lines, calculation of
force coefficients on bodies, and generation of elements for visual post-
processing, such as section planes and iso-surfaces [46]. Furthermore, writing
only the data of interest requires a much smaller memory footprint than the full
dataset, reducing the time required to post-process a solution as well as the
overall disk space requirements for a time-accurate solution.
3.5.6 Solution Monitoring
Monitoring solution progress is important to ensuring computational
accuracy and determining solution convergence. During execution of a general
simulation, OpenFOAM outputs convergence residual statistics as well as local
and global Courant number information for each iterative solution process. This
output can be sent to a log file using the foamJob script file to initiate the solver
as a background process; the log file can then be monitored in real-time with the
tail UNIX command. The foamLog script file processes the log file and
generates files corresponding to each monitored residual for graphical
comparison.

31
3.5.7 Additional Community-developed Tools
As seen in the previous sections, OpenFOAM contains a great deal of
functional tools already built into the existing code database. The customization
offered by the open-source code has also allowed users within the community to
develop their own tools and distribute them separate from the original CFD
package. Two prominent examples of useful user contributions are PyFoam and
swak4Foam [47]. PyFoam is a Python library that has been developed to control
OpenFOAM simulations and manipulate data in a command line fashion using
Python commands. It includes utilities for advanced solution monitoring, clearing
and duplication of case files, and manipulating dictionary files and field definitions
to quickly modify case settings. The swak4Foam (Swiss Army Knife for FOAM)
library expands beyond the basic setFields utility by enabling the user to enter
non-uniform and analytical boundary conditions and field values using parsed
complex mathematical expressions, including variables and conditional
statements. Both of these tools are used by a significant portion of the
OpenFOAM community to support analysis and execution of simulations and
have been used in this thesis to assist with setup and monitoring tasks for the
simulations in Chapter 5. The simulation monitoring and input file modification
capabilities provided by these tools are necessary to ensure efficiency using
OpenFOAM in an engineering production environment.

32
CHAPTER 4
SOLUTION METHODOLOGY

In this chapter, an overview of the methodology used in developing the
Vorticity Confinement solver in OpenFOAM is presented. In actuality two
separate solvers have been developed in the current work to stay true to the
component-based approach used in OpenFOAM; pisoVcFoam is a time-
accurate, implicit time-stepping solver based on the PISO (Pressure-Implicit with
Splitting of Operators) algorithm, while simpleVcFoam is a steady-state iterative
solver based on the SIMPLE (Semi Implicit Method for Pressure Linked
Equations) algorithm. Both numerical algorithms have been standardized in most
of the pressure-based OpenFOAM solvers included in the CFD package; the
inclusion of the VC1 and VC2 methodologies within the PISO and SIMPLE
frameworks, however, is a main development of this work.
The finite volume formulation of the Navier-Stokes equations and its
treatment in OpenFOAM is described in Section 4.1. Because OpenFOAM uses
a collocated (non-staggered) mesh arrangement, a typical Rhie-Chow correction
is needed to couple the velocity and pressure solutions with the PISO and
SIMPLE operator-splitting algorithms. OpenFOAM does not perform this
correction explicitly, but instead uses a similar process to correct the solution
which can be described as in the spirit of Rhie-Chow [48]. The Rhie-Chow
implementation and the details of the PISO and SIMPLE algorithms are
presented in Section 4.2. The implementation of the Vorticity Confinement model
is similar for both solvers and is discussed in Section 4.3. Additionally, the
immersed boundary treatment of no-slip boundaries within the flow field is
presented in Section 4.3.2.

33
4.1 Finite Volume Approach
There exist three typical discretization approaches for the numerical
solution of PDEs: finite difference, finite element, and finite volume. Each method
requires solving over a grid of points defining the domain of interest; the
differences between approaches arise in the solution methodology. Short
descriptions of the finite difference and finite element methodologies are given
here in order to highlight the benefits of the finite volume approach.
The finite difference approach solves the conservation equations in
differential form by approximating the partial derivatives at each grid point in
terms of the surrounding nodal values, creating an algebraic equation at each
node [1]. Its main advantage is the ease in obtaining higher-order approximations
using Taylor series expansions; however, conservation is not explicitly enforced
with finite difference equations and requires special treatment to ensure a
conservative solution [49].
The finite element approach is similar to finite volume in that it is based on
the integral formulation of the conservation equations; however, with finite
element methods the conservation equations are multiplied by a weighting
function based on a linear shape function within each element to ensure
continuity between elements. The equations are then solved by requiring that
derivatives at each nodal value equal zero, which results in a solution based on
minimization of the shape function residuals [1].
To contrast with the aforementioned solution methodologies, the
discretization approach used within OpenFOAM and the current work is the finite
volume method, which is based on the integral form of the conservation
equations. The solution domain is subdivided into control volumes with the
solution points defined at the centroids of each control volume. Values from the
centroids are interpolated to the surfaces (faces) of the control volumes. Surface
and volume integrals can be evaluated from the values on the faces and an
algebraic equation can be obtained for each control volume, which also has the
benefit of ensuring explicit conservation [1]. The integral form of the conservation

34
equations also allows for the presence of discontinuities, such as shocks, within
control volumes, whereas the differential form used for finite difference methods
requires special treatment for discontinuities because it assumes that flow
properties are continuous [50]. The main disadvantage of finite volume methods
is the difficulty extending the schemes to higher than second order due to the
interpolation requirements for control volume surface values [1]. For most
engineering calculations, however, second-order accuracy is sufficient to provide
an accurate result and higher order accuracy becomes illusory in the presence of
thin vortical structures or turbulence.
4.1.1 Conservation Equations in Integral Form
The integral conservation laws governing an isothermal, constant density,
incompressible fluid for a control volume fixed in space can be written as follows:

(4.1)

(4.2)

where U is the velocity vector and f includes body and gravitational forces. The
stress tensor, , includes pressure and viscous stresses and is given as follows:

( ()
) (4.3)

Equation (4.2) is a mix of surface and volume integrals which makes it
somewhat difficult to evaluate numerically. To manipulate the surface integrals,
the generalized form of Gauss theorem produces the following useful identities
[48]:

35

(4.4)

(4.5)

(4.6)

S represents the bounding surface of the control volume, where dS is a
surface element on S with a surface normal n directed outwards from the control
volume. Tensor quantities are represented by a, while scalar quantities are
described by . Using the identities in Equations (4.4)-(4.6), the surface integrals
in Equations (4.1) and (4.2) can be converted to volume integrals to be evaluated
across the control volumes and the conservation equations can be written
symbolically in differential form:

(4.7)

() ()
(4.8)

To simplify the discussion, the terms in Equation (4.8) can be arranged
into Equation (4.9) according to generic discretization treatment within the finite
volume approach. Terms requiring explicit evaluation such as the contribution of
pressure and external sources are placed on the right hand side of the equation.
Terms on the left hand side of the equation can be treated using implicit
evaluation which allows for a more stable calculation. The terms in the equation
are, from left to right: the temporal derivative, convection term, diffusion term,
contribution of the pressure, and the explicit source term.

() ()
(4.9)

36

In the case studies presented in Chapter 5, the temporal derivative is
discretized using either an implicit backward Euler time scheme or an implicit
Crank-Nicholson method, both of which are second-order accurate in time. A full
explanation of the backward Euler and Crank-Nicholson time scheme
implementations for OpenFOAM, and hence the finite volume approach, is
provided in Hrvoje Jasaks thesis [48] and is not discussed here further.
An explanation of the spatial term discretization is necessary to
understand the implementation of the Rhie-Chow correction, discussed in
Section 4.2.3, which is necessary for accurate pressure-velocity coupling on the
OpenFOAM collocated mesh structure. A discretized form of Equation (4.4) is
derived by assuming linear (second-order) variation across the control volume
and assuming that the faces of the control volume are flat surfaces based on the
mesh quality criteria detailed in Section 3.3 [48]:

( )
(4.10)

In this discretized form, Vp represents the volume of a discrete cell with
centroid point P. The f label denotes a face value of the quantity of interest, while
S represents the discrete surface area of the face. Applying this form of Gauss
theorem in Equation (4.10) to the convection term recovers the following:

()
()

()
(4.11)

37
In the final manipulation of Equation (4.11), F represents the mass flux
through each face, and Uf is the velocity at the face interpolated from the control
volume centroid. The flux must satisfy continuity in Equation (4.7) which creates
a non-linear relationship. To linearize the convection term, the flux contribution
(and resulting non-linearity) is lagged and uses the existing velocity field to
calculate the mass flux. This requires iteration over the non-linear terms to
accurately resolve the non-linear effects which will be discussed in Section 4.2.
The diffusion term is discretized in a similar manner to the convection
term. Applying Equation (4.10) to the diffusion term and assuming a linear
variation across the control volume results in the following:

()
()
()
(4.12)

The gradient of velocity at each face is interpolated based on the values of
velocity at the two unique cell centers on either side of each face. The most
accurate interpolation occurs when the mesh is orthogonal and the face normal
vector is parallel to the vector between cell centroids; if the mesh is non-
orthogonal, correction factors are necessary to maintain sufficient accuracy or
the solution may become unbounded.
4.2 Pressure-Correction Methods
Pressure-correction methods form the basis for the solution of the Navier-
Stokes equations in OpenFOAM for incompressible flow. Pressure-correction
methods were first introduced by Chorin [51] for incompressible flow regimes and
are a subset of projection methods using the operator-splitting approach. The
solution is advanced by a predictor step in which the momentum equation is
solved for a divergent intermediate velocity field. The algorithms rely on the

38
assumption that the intermediate known velocity field can be decomposed into
unknown divergence-free (solenoidal) and irrotational parts [52]:

(4.13)

The irrotational part can also be represented by the gradient of a potential
shown in (4.13). Enforcing continuity requires the divergence-free contribution
equal zero by definition and recovers a Poisson equation for the unknown
potential based on the known velocity field.

(4.14)

After solving for the potential, the new divergence-free velocity field is
calculated by rearranging (4.13) and correcting the previous velocity field with the
known potential to advance the solution to the next time step or iteration:

(4.15)

Solving for the velocity field using a pressure-correction method requires
assembling a system of equations at each solution point in the domain. For each
cell, all temporal and spatial discretization operations in Equation (4.9) can be
condensed into the following algebraic equation [48]:

(4.16)

To solve Equation (4.16) at the current time level n for the value at the cell
centroid P requires the values of the neighboring N cells at time level n and the
explicit source term,
. Each control volume is associated with a coefficient

and neighboring control volume coefficients
. The solution requires the

39
assembly of the entire system of equations for all control volumes in block-matrix
form:

[][] [] (4.17)

In this form, the A matrix is a sparse matrix constructed of diagonal
coefficients
and off-diagonal coefficients
, the U vector matrix has one

element per control volume, and the r vector matrix holds the source term for
each control volume. Using this terminology enables tensor operations spanning
the entire mesh to be written with symbolic equations. The pressure-correction
approach requires the derivation of an explicit pressure equation which begins
with the solution of a predictor step written in Equation (4.18):

(4.18)

In this step, the old and new time levels are designated by o and n,
respectively. Depending on the selected temporal discretization there could
potentially be additional time levels; however, for derivation purposes only one is
shown here. The a coefficients encapsulate necessary dimensional details such
as the cell volume and time step. To express the equation as a solution for UP, an
additional term H(U) is defined which holds the off-diagonal contributions as well
as any additional source terms [48]:

()

(4.19)

()
(4.20)

40
Enforcing continuity in Equation (4.7) on Equation (4.20) allows for the
derivation of the pressure equation. The right hand side is transformed by
applying the identity in Equation (4.10):

(
()
)
(
)
()
) (
()
(4.21)

The final discretized system of equations requires the solution of the
Poisson system in Equation (4.21) in a similar manner to the diffusion term in
Equation (4.12):

(
()
(())

(4.22)

To calculate the velocity for the new time step, Equation (4.20) is
transformed by applying the identity in Equation (4.5) to the pressure term, which
allows the use of the solved pressure from Equation (4.22) on the faces:

()
()

(4.23)

This completes the iteration and the calculation can either continue
iterating or move on to the next time level, depending on the convergence
criterion. The following sections describe the time-accurate and steady-state
pressure-correction algorithms used in OpenFOAM.

41
4.2.1 PISO Algorithm for Time-Accurate Simulation
The PISO (Pressure-Implicit with Splitting of Operators) algorithm is an
implicit time-marching scheme first developed by Issa which uses operator
splitting to separate the solution of the velocity and pressure equations, such that
the solutions of each field are close approximations of the exact solution with
formal second-order temporal accuracy [53]. The solution procedure is as
follows:
1. Solve the discretized momentum equation implicitly for the predictor
velocity using the current values of velocity and the pressure. For the
first iteration the pressure from the previous time step is used.
2. Compute the mass fluxes based on the predictor velocity.
3. Assemble and solve the pressure-correction equation for the corrective
pressure using the predictor velocity field.
4. Correct the mass fluxes to be divergence-free. This divergence-free
field is used in the discretized convection term in the next time step.
5. Correct the predictor velocity explicitly to obtain a divergence-free
velocity field.
6. Repeat steps 2-5 using the new velocity field as a predictor velocity for
a set number of iterations. Issa recommends a minimum of two
predictor-corrector iterations to achieve an accurate solution [53].
7. Update boundary conditions, increase the time step and repeat from
step 1.

A flow chart representation of the PISO algorithm compared with the
SIMPLE algorithm (discussed in Section 4.2.2) is shown in Figure 5. The main
differences between SIMPLE and PISO are the lack of relaxation and the
additional iteration on the pressure equation in PISO. Issa [53] provides evidence
that the errors associated with the method due to operator splitting are small
enough that the algorithm does not require iterative solution of the momentum
equation.

42

Figure 5. Flow chart comparison of PISO and SIMPLE algorithms.

43
Favorable results for both time-accurate incompressible and subsonic
compressible flows were presented by Issa et al. [54] in which it was
demonstrated that PISO is also stable for larger time steps approaching pseudo-
steady simulation.
4.2.2 SIMPLE Algorithm for Steady-State Simulation
The SIMPLE (Semi Implicit Method for Pressure Linked Equations)
algorithm was introduced by Patankar [55] for incompressible flow simulation.
While it is capable of solving time-accurate flow problems, other methods (such
as PISO) are more efficient and as a result the SIMPLE algorithm has popularity
primarily with steady-state simulation. The solution procedure in OpenFOAM is
as follows:
1. Solve the discretized momentum equation with a relaxed velocity field
and the pressure field from the previous time step to compute an
intermediate velocity field.
2. Compute the mass fluxes based on the intermediate velocity.
3. Solve the pressure equation and apply under-relaxation to the
pressure field.
4. Correct the mass fluxes to be divergence-free. This field is used in the
discretized convection term in the next time step.
5. Correct the velocity field using the new pressure field.
6. Update the boundary conditions and repeat with step 1 until
convergence.

On a per-iteration basis the SIMPLE algorithm is faster than PISO
because it is solving the pressure equation only once; however, in a time-
accurate calculation the SIMPLE algorithm must apply several iterations per time
step to obtain an accurate solution. The increased iterations result in a longer
calculation time per time step compared to the PISO algorithm.

44
4.2.3 Rhie-Chow Treatment in OpenFOAM
In the past, the use of pressure-correction algorithms on collocated
meshes has exhibited decoupling of the velocity and pressure fields and
subsequent oscillations within the solution. Normal central differencing on
collocated grids causes a checkerboard effect in which alternate grid points are
skipped in the differencing operation which leads to the oscillations [49]. Rhie
and Chow [56] proposed a collocated grid arrangement which counters this
effect. In their formulation, the momentum equation is solved for velocity as in
Step 1 of the PISO and SIMPLE algorithms. An intermediate velocity from which
the pressure contribution is explicitly removed is then interpolated from the cell
nodes to the cell faces. To find the corrected velocity at the faces, the pressure
gradient is then added back to the interpolated velocity; however, the important
point is the pressure gradient is evaluated at the faces of each cell rather than
calculated from opposing cell nodes. This correction is effectively a higher-order
pressure correction which eliminates the oscillations and satisfies continuity.
OpenFOAM does not have such an explicit correction factor, but as previously
stated employs a methodology described as in the spirit of Rhie-Chow [48].
Figure 6 displays a high-level C++ code snippet which details the PISO
algorithm including the implicit Rhie-Chow correction; while the code is fairly
understandable purely from the syntax, each operation will be explained for
clarity. Lines 1-6 initialize the coefficients of a finite volume matrix UEqn created
from the temporal, convection, and diffusion terms. The momentum equation
predictor using the UEqn object along with the gradient of the current pressure
field is solved for an intermediate velocity in line 8; the for loop in lines 9-24 is
the iterative pressure corrector step. Two functions UEqn.A() and UEqn.H() are
called on lines 11-12, which are equivalent to the
and H(U) terms in Equation

(4.20). Line 12 is the explicit removal of the pressure from the momentum
predictor for the Rhie-Chow correction; the H(U) object in Equation (4.19) does
not include the pressure. Because the momentum predictor is implicit this result
is different than simply solving Equation (4.18) without the pressure entirely.

45
Following the Rhie-Chow methodology, this velocity is interpolated to the
faces in line 14; solving for the flux at each face involves a dot product between
the normal mesh surface and the interpolated velocity field. Lines 16-20 describe
the solution of the pressure equation which is equivalent to Equation (4.21). The
mass flux is corrected on line 21. Line 22 completes the implicit Rhie-Chow
correction by correcting the velocity without a pressure contribution from line 12
with the gradient of pressure based on the face values. To summarize, Rhie-
Chow is implemented implicitly in OpenFOAM by removing the effect of pressure
from the flux phi, solving the divergence of the pressure term in Equation (4.21)
using the gradient of pressure on the cell face, and correcting the predictor
velocity using the pressure gradient calculated from the cell face values.
4.3 Vorticity Confinement Implementation
The implementation of Vorticity Confinement for the finite volume
approach follows the methodology suggested for compressible flows, in which

Figure 6. OpenFOAM C++ code snippet for the PISO algorithm.

46
the VC term is introduced to the integral equations as a body force source term.
As stated in Chapter 2, the implementation used within this work follows the
suggestion of Costes [28] in which the VC term is left out of the energy equation
and added to the momentum equation alone. This treatment is appropriate
because the VC term has no true physical representation and is purely a
numerical treatment for vortical regions. An identical implementation approach is
used for both VC1 and VC2 formulations which takes advantage of similarities
between the two VC methods, with the difference occurring in the formulation of
the source term as outlined in Section 2.1 and 2.2.
The similarities between the PISO and SIMPLE algorithm flowcharts
displayed in Figure 5 suggest that a general Vorticity Confinement approach for
pressure-correction methods can be developed. One concern with these
methodologies, however, is the treatment of non-linear source terms. Equation
(4.19) indicates that the source term is placed in the H(U) operator, which is
acceptable for implicit source terms because the H(U) operator is reevaluated
based on the current velocity field at the beginning of each pressure correction
step. For explicit source terms, however, any non-linear behavior in the source
term is lagged and is calculated using the previous time step velocity field when
the term is initialized. This treatment is somewhat problematic for Vorticity
Confinement; the intention of the additional term is to provide anti-diffusion in
vortical regions and in the original formulation by Steinhoff [2] the term is
calculated using the current time step velocity field. Lagging the VC term results
in anti-diffusion based on the previous time step which may not be optimal for the
calculation.
To address this issue a novel formulation has been developed for
pressure-correction methods which allows the VC term to be calculated based on
the velocity field originating from the momentum predictor step rather than the
previous time step velocity. This approach is in the spirit of the original VC
approach which uses a fractional-step solution methodology where each operator
is solved sequentially. The algorithm begins by solving for the momentum

47
predictor without the VC term using Equation (4.18) as in the normal pressure-
correction method:

The off-diagonal terms and source terms, with the exception of the
pressure gradient, are collected into the H(U) operator as shown in Equation
(4.19). The VC term is then introduced in a modified version of Equation (4.20):

()
(4.24)

The VC term is calculated before Equation (4.24) based on the resulting
intermediate velocity from the momentum predictor solution of Ux, Uy, and Uz.
Because the VC term is a numerical correction rather than a physical term,
inserting it at this point in the solution process causes no ill effects to solution
stability. Effectively the VC term is correcting the momentum predictor before
proceeding to the pressure corrector. The solution algorithm progresses through
the steps outlined in Equations (4.21) through (4.23), with the VC term joining
with the H(U) operator term in the equations. In this manner the VC correction is
calculated based on the current time step velocities and avoids the lagging
behavior of the explicit source term. Results presented in Chapter 5 demonstrate
the viability of the implementation.
4.3.1 Parallelization
Parallel computation is critical for present-day CFD engineering
applications. The size and scope of industrial problems necessitates using
millions of cells to resolve the flow to an acceptable level of accuracy, which in
turn requires solving on multiple processors to obtain a solution in a reasonable
amount of physical time. To address this requirement, one of the goals of the

48
current work mentioned in the introduction is the development of a fully
parallelized solver.
Parallelization is implemented in OpenFOAM from lower-level classes,
which implies that most user-developed applications require no specific coding to
realize parallel operation. Most of the operations in the calculation of the VC
source terms involve manipulation of fields spanning the entire mesh using
tensor mathematics. These operations are inherently parallel from the lower-level
classes. Manipulation of fields on a cell-by-cell basis, however, requires specific
coding to ensure proper communication between processor domains. In the VC2
formulation, the harmonic mean of vorticity is used which must be calculated over
each cell. An efficient algorithm has been developed to calculate the harmonic
mean in Equation (2.8) and is detailed in Figure 7. Lines 1-4 create the fields
used in the calculation; the SMALL constant is used here to prevent divide by zero
errors. The forAll loop on line 5 orders all processors to iterate over the cells in

Figure 7. C++ code snippet for harmonic mean algorithm.

49
each decomposed portion of the mesh; a list of neighboring cells is created on
line 6 and is iterated in lines 7-9 for a summation of the reciprocal term. The
contribution of the current cell is included in lines 10-11.
The forAll loop in lines 13-23 encompasses the portion of the harmonic
mean which requires inter-processor communication. In the decomposition
process, the processor boundaries are marked with a particular type of boundary
condition. Extracting the coupled boundaries on line 16 allows for values from
the neighboring processor to be retrieved using the patchNeighbourField()
method. The harmonic mean on processor boundary cells requires these values
for proper calculation. On line 24, the count and sum fields, which have been
holding the number of cells in each stencil and the summation of the reciprocal of
vorticity over the stencil, can be multiplied with the vorticity as a field operation to
create the W term in Equation (2.7) for the VC2 formulation. This completes the
parallel algorithm for computing the harmonic mean.
4.3.2 VC as an Immersed Boundary Method
Immersed boundary methods are alternative methods for viscous flow
simulation in which boundaries are embedded within non-conforming grids, as
opposed to conventional solution approaches using body-fitted grids. The basic
approach involves modifying the equations for cells in the vicinity of the
immersed boundary to take into account the desired boundary conditions; in this
manner, a generic grid of specific resolution can be used for different geometry
configurations without requiring repeated grid generation. Mittal and Iaccarino
[57] provide a full overview of conventional immersed boundary methods and
discuss both advantages and disadvantages of the various methodologies.
Conventional methods involve two different formulations, namely the continuous
forcing approach and the discrete forcing approach. The continuous forcing
approach involving the application of an analytical forcing function was first
suggested by Peskin [58] for dynamic heart simulation and has since been
extended to rigid body simulation [59]. Discrete forcing methods are the focus of

50
recent research and involve extracting a discrete forcing function directly from the
numerical solution of surrounding cells. These methods include the indirect
forcing approach of Mohd-Yosuf [60], the ghost-cell approach [61,62,63], and the
cut-cell approach [64].
In the current work, an immersed boundary method is implemented similar
to that proposed by Fan and Steinhoff [6] and employed by Lynn [24]. The
previous formulations utilize a level set function which is essentially a signed
distance field marking cells inside and outside the body. During the calculation,
the solution algorithm proceeds through all cells and velocities in interior cells are
simply set to zero. The current effort uses a customized pre-processing tool
specifically developed to mark inside and outside cells as separate sets. The
set identification for each cell is determined by comparing its location to the
surface normal of a watertight CAD surface embedded within a background
volume mesh; after all cells have been separated between the two sets, the sets
can be converted to zones which are automatically decomposed during a
parallel computation. As in the aforementioned formulations, the velocity in
interior cells is set to zero during the calculation to generate the immersed
boundary. This process is rather crude numerically and without VC provides only
a first-order approximation of the boundary with considerably higher
discretization error and staircase effects along the immersed boundary
interface. Applying the unique VC formulation compresses the vorticity towards
the immersed boundary, containing the discretization error within the boundary
layer, and diffuses vorticity along the boundary, smoothing any staircase
behavior along the immersed boundary interface allowing a sharp boundary
definition to emerge. In Chapter 5, the immersed boundary method is verified in
5.2 with an immersed slanted plate simulation and shown to be suitable for
complex geometry simulation in 5.5 with the time-accurate simulation of an
immersed Formula One racecar at yaw.

51
CHAPTER 5
RESULTS AND DISCUSSION

This chapter presents an overview of the results of three two-dimensional
verification cases as well as two applications demonstrating complex three-
dimensional simulation. The primary focus of these simulations is to establish
and verify the implementation of Vorticity Confinement presented within this
thesis and quantify the effects of the VC methodology on the resulting flow field.
According to the AIAA Standards Guide for the Verification and Validation of
Computational Fluid Dynamics Simulations [65], verification is defined as the
process of determining that a model implementation accurately represents the
developer's conceptual description of the model and the solution to the model. To
demonstrate verification of the developed VC algorithm, results from the
verification cases are shown with and without VC and the effects of the method
on the resulting flow field are discussed in context with results shown by
Steinhoff and others in similar studies [4,15,17,18].
Each of the presented case studies utilizes either of the developed
solvers, pisoVcFoam (time-accurate solver) or simpleVcFoam (steady-state
solver), depending on the nature of the temporal approximation. Discretization of
the governing equations follows standard Gaussian integration described in
Chapter 4, with the cell-to-face interpolation scheme determining the order of
accuracy of the solution. Both solvers use a second-order upwind scheme for
interpolation of the convection term and a second-order central differencing
scheme for interpolation of the diffusion term. Two separate discretization
schemes are used for the temporal term in the pisoVcFoam solver. For cases in
5.3 and 5.5, an implicit second-order Crank Nicholson scheme is used which
takes the mean of the n-1 and n solution values. This scheme requires only two
time levels and is unconditionally stable but can become unbounded and
produce a false solution if the time step is too large. An implicit second-order

52
backward differencing scheme is used in the 5.4 case study to match temporal
discretization used in the large-eddy simulation (LES) study provided for
comparison. This scheme uses three time levels (n, n-1, and n-2 solution values)
and ensures boundedness, but requires significantly more memory overhead.
5.1 Flow over a Flat Plate
This verification case involves flow traveling over a two-dimensional grid-
aligned flat plate with a no-slip boundary condition. A boundary layer is
generated due to the no-slip wall condition and the thickness can increase in
height as the flow progresses downstream along the plate. Applying Vorticity
Confinement as a boundary layer model has shown in previous studies [7,24]
that with increasing values of , VC reduces the boundary layer thickness to 2-3
grid cells by providing a compressive effect on the vorticity in the boundary
normal direction. This simple case provides an opportunity to calibrate the
confinement strength parameter to an appropriate value which can be used as a
starting point for future cases.
5.1.1 Setup Details
The two-dimensional computational mesh used in this verification case is
shown in Figure 8. The domain is 2.5 m x 2.5 m and is discretized by a 0.01 m
uniform mesh resulting in 62,500 cells. Although a problem of this size is easily
solvable using a serial process, the mesh has been decomposed into four equal
subdomains using simple geometric decomposition to demonstrate the parallel
capability of the developed VC solver. The flow is initialized with a uniform
velocity of (1, 0) where the x-axis is parallel to the plate and the y-axis is normal
to the plate; flow moves from left to right. The applied boundary conditions are
shown in Figure 9. A slip wall is used for the first 0.5 m away from the uniform
inlet to avoid spurious boundary interactions. A no-slip wall continues for 2 m to
the domain outlet. A constant pressure boundary condition is applied at the outlet
and Neumann boundary conditions are used at all other boundaries for pressure.

53

Figure 8. Computational mesh for flat plate case study. Processor subdomains have been
separately colored to show domain decomposition.

Figure 9. Boundary conditions for flat plate case study.

54
For the velocity field, a prescribed uniform velocity is used at the inlet
boundary and a freestream boundary condition with flow parallel to the inlet
velocity is set for the far field boundary. The steady-state solver simpleVcFoam is
used because only the steady-state nature of the boundary is of interest for this
case. Isotropic numerical diffusion is explicitly applied using a value of = 0.001
m
2
/s. It is necessary with VC to impose isotropic diffusion on the vortical features
of the flow which can then be counteracted by the isotropic compression imposed
with the calibrated constant confinement strength [66]. Using the aforementioned
settings, a baseline case without VC has been developed and is compared
against three VC1 cases and three VC2 cases. The VC1 cases utilize a
combination of Field Confinement and Surface Confinement and apply the same
confinement strength value to both formulations. The VC2 formulation does not
require any special treatment near the boundary. The confinement strength is
varied in each VC case; in this manner, a range of confinement parameters can
be calibrated for each formulation and the effects of the VC model on the flow
field can be observed.
5.1.2 Baseline Results
The velocity magnitude and vorticity contours for the baseline case without
VC are shown in Figure 10 and Figure 11. On all contour plots the domain has
been clipped with the upper half hidden; this portion of the domain is only
necessary to prevent unnatural constraints on the pressure solution and does not
add value to the boundary layer solution. Based on the velocity distribution in
Figure 10, the boundary layer begins developing at the no-slip wall and continues
growing until it reaches the outlet on the right side of the domain. The vorticity in
Figure 11 has the highest value at the initial onset of the no-slip boundary and
then diffuses outward away from the wall. An effective low Reynolds number
solution is achieved due to the high numerical diffusion.

55

Figure 10. Velocity magnitude contour for flat plate case without VC.

Figure 11. Vorticity magnitude contour for flat plate case without VC.

56

Figure 12. Boundary layer profile of flat plate case without VC at various locations along
the plate. (x) refers to the distance from the leading edge.
A plot of the boundary layer profile in Figure 12 at various distances from
the leading edge of the plate shows the development of the boundary layer along
the plate. The streamwise velocity (x-axis) has been normalized by the local
freestream value at each station along the plate, and distance from the wall (y-
axis) has been normalized by the uniform grid cell size. Based on this correlation,
at the 1.875 m station the height of the boundary layer has reached
approximately 20 cells. Without a turbulence model the boundary layer assumes
a laminar boundary layer shape, growing much faster and becoming much
thicker than a turbulent boundary layer.
5.1.3 Effect of VC on Boundary Layer
A typical near-wall boundary layer solution for conventional CFD uses wall
functions which prescribe a velocity distribution based on experimental and
analytical correlations; however, being based on average quantities this type of

57
model has difficulties predicting separation and other turbulent phenomena.
Vorticity Confinement provides a means to overcome this difficulty by providing a
framework in which the boundary layer relaxes to a defined shape but is still
allowed to react to separation-inducing effects. VC compresses the boundary
layer to the height of 2-3 grid cells and relaxes to a thin state. The solution within
the boundary layer region is assumed to be unimportant to the macroscopic flow
field; the localized effect of keeping the boundary layer thin simulates the shape
of the turbulent boundary layer.
Results for the boundary layer modeled with varying parameters of VC1
and VC2 are shown for the 1.875 m downstream station in Figure 13 and Figure
14. In both formulations as the confinement strength is increased the resulting
boundary layer height decreases, approaching the 1/7 power law relationship of
a turbulent boundary layer. Increasing the confinement strength to a suitably high
value confirms that the boundary layer can be reduced to the thickness of only a
few grid cells, overcoming the explicit numerical diffusion as well as any
numerical diffusion inherent within the solver. This boundary layer shape is held
constant along the length of the plate to approximate the shape of the 1/7 power
law relationship by keeping the boundary layer thin. Figure 15 and Figure 16
depict the boundary layer profiles for the VC1 and VC2 simulations with the
highest established confinement strength values. For both formulations, the
boundary layer relaxes to a corresponding thickness and remains at that height
along the remainder of the plate. In this case there are no perturbations in the
flow which disturb this behavior. In a three-dimensional turbulent environment,
however, this thickness will vary based on the surrounding flow effects and can
allow separation as necessary in response to an adverse pressure gradient but
will revert back to the relaxed state when unperturbed.
Figure 17 and Figure 18 depict velocity magnitude and vorticity magnitude
contours for both the VC1 and VC2 simulations with the highest established
confinement strength values. Both results are exceedingly similar; the only
difference which stands out is a slight increase in velocity just above the

58

Figure 13. Varying VC1 confinement strength and its effect on boundary layer height at
downstream station x = 1.875 m.

Figure 14. Varying VC2 confinement strength and its effect on boundary layer height at
downstream station x = 1.875 m.

59

Figure 15. Boundary layer profile from the flat plate simulation with VC1 confinement
strength =0.25.

Figure 16. Boundary layer profile from the flat plate simulation with VC2 confinement
strength =0.004.

60

Figure 17. Velocity magnitude contours for VC1 (top) and VC2 (bottom) simulations with
confinement strength values of 0.25 (VC1) and 0.004 (VC2). Note that a constant diffusion
() of 0.001 is applied.

61

Figure 18. Vorticity magnitude contours for VC1 (top) and VC2 (bottom) simulations with
confinement strength values of 0.25 (VC1) and 0.004 (VC2). Note that a constant diffusion
() of 0.001 is applied.

62
boundary layer in the VC2 simulation which is evident in the boundary layer
profile in Figure 16. The slight acceleration is due to the momentum conserving
aspects of the VC2 formulation which will be visited in more detail in 5.3. This
effect is comparable to the accuracy provided by many other boundary layer
models and is of little concern.
For VC1 the established stable confinement strength parameter range is
between 0.01 and 0.25, while VC2 shows stable behavior between 0.001 and
0.004. Values above this range for the flat plate boundary layer result in a chaotic
flow field with large-scale enhancement and excitation of vortical structures. The
investigated confinement strength ranges for VC1 and VC2 are comparable to
values reported in previous studies [18]. In particular, the stable values for the
VC2 formulation lie within the approximate range suggested by Steinhoff et al. of
1 4 [67]. The agreement of the confinement parameter with previous
studies and the demonstrated behavior of the solver provide confidence that the
method has been implemented correctly using the OpenFOAM framework.
5.2 Flow over a Slanted Plate
The case in 5.1 demonstrates the ability of Vorticity Confinement to
confine the boundary layer to a region with a thickness of 2-3 grid cells. The
ability to confine the vorticity into a thin sheet can be extended to produce an
immersed boundary method using the VC model. Using the process described in
4.3.2 to implement the immersed boundary, a sheet of vorticity is implicitly
generated at the interface between the inside cells being set to zero velocity and
the outside cells containing the flow solution. The VC model is then able to
confine this vorticity sheet back to the immersed surface (point of highest
vorticity) and provide an accurate representation of surface effects without a
body-fitted grid. Defining the surface irrespective of the surrounding fluid grid
cells has advantages for rapid prototyping, such that the surface can be easily
changed while the surrounding mesh remains unaltered. This allows many

63
similar simulations to be run in a short amount of time and eliminates the manual
interaction required to generate a new volume mesh.
5.2.1 Setup Details
The 5.2 case study involves the simulation of an immersed flat plate at a
15 angle with the base of the domain. This case utilizes the identical mesh from
the 5.1 case study as well as the steady-state solver simpleVcFoam and
therefore the details of the simulation setup described in 5.1.1 apply to the 5.2
setup as well. Identical details will not be repeated but the differences between
the two simulations will be highlighted. Figure 19 displays the boundary
conditions used for the immersed flat plate. The immersed surface is represented
by a discrete CAD surface and is shown superimposed on the background mesh;
the velocity components for each cell within this body are set to zero at each step
during the iterative process. The inlet flow direction is set parallel to the

Figure 19. Boundary conditions for slanted plate case study. Immersed body is shown
superimposed on the mesh.

64
immersed surface to reduce upstream effects from the no-slip immersed
boundary. Due to the angled inlet flow direction, a constant pressure boundary
condition is used at the top of the domain instead of the freestream boundary
used in the 5.1 case study. All other boundaries remain similar to the 5.1 setup.
5.2.2 Immersed Boundary with VC1
An initial study is performed using a simple laminar calculation (no
turbulence model) without the contribution of VC. The results are displayed in
Figure 20 and Figure 21 detailing the velocity and vorticity magnitude contours.
The contours are shown with values for all cells, including those found within the
immersed body. The blue region with zero velocity in Figure 20 is within the
immersed body. Setting the velocity to zero effectively creates a no-slip wall
condition at the interface between the immersed region and the flow region,
generating a boundary layer which diffuses away from the immersed surface.
This boundary layer also includes staircase effects, abnormalities within the flow
which are aligned to the grid cell shapes near the immersed surface and due to
the step function approach of the immersed boundary method. This is slightly
evident in the vorticity contour in Figure 21; however, most of the staircase
effects in this problem are smoothed by the numerical diffusion. Conventional
immersed boundary methods treat this effect with a variety of ad-hoc
interpolation methods which prescribe a velocity in the cells near the immersed
surface. Vorticity Confinement, on the other hand, requires no special treatment
and instead convects the vorticity directly back to the immersed surface, ensuring
a smooth boundary shape. Velocity and vorticity magnitude contours for the VC1
formulation are found in Figure 22 and Figure 23. With the VC1 model, a smooth
boundary layer is developed over the slanted plate which remains at a fixed
height and is distributed over only a few grid cells at the immersed surface. The
maximum value established for the VC1 confinement strength parameter in the
flat plate boundary layer case is used for both field and surface confinement and
proves to be sufficient to fully confine the boundary layer to a thin sheet.

65

Figure 20. Velocity magnitude contour of immersed plate without VC.

Figure 21. Vorticity magnitude contour of immersed plate without VC.

66

Figure 22. Velocity magnitude contour for immersed plate with VC1.

Figure 23. Vorticity magnitude contour for immersed plate with VC1.

67
5.2.3 Immersed Boundary with VC2
The velocity and vorticity magnitude contours for the VC2 formulation
using the maximum established confinement strength from 5.1 are shown in
Figure 24 and Figure 25, respectively. The solution exhibits boundary layer roll-
up near the end of the plate, indicated by the vortices forming within the
boundary layer region. Such transient behavior is the result of combining a
second-order numerical scheme for the convection term with the highest
confinement strength value which minimizes the numerical diffusion and allows
the appearance of transient instabilities. To test this relationship, an additional
case using the same VC2 confinement parameters and first-order upwind
differencing is developed. The velocity magnitude contour displayed in Figure 26
illustrates that the additional numerical diffusion added by the first-order upwind
differencing eliminates the transient behavior and causes the boundary layer to
thicken. A reduction in the VC2 confinement strength should likewise effectively
increase the numerical diffusion within the simulation and show a similar
response. Figure 27 details the velocity magnitude contour of a VC2 simulation
using the original second-order convection scheme with confinement strength of
0.0035 (the previous value was 0.004). The boundary layer instability no longer
appears in the solution but the boundary layer remains thin as the confinement
strength was only reduced by 12.5%. The sensitivity of the VC2 model to the
variation in numerical schemes confirms the findings of Butsuntorn and Jameson
[32] that the confinement strength value is dependent on the numerical scheme
being used. Using at least a second-order scheme serves to reduce the
discretization error present and minimize the effect of numerical diffusion; as a
result, the variance of the confinement parameter between simulations with a
higher-order scheme is expected to be much less than with lower-order
schemes. The transient instabilities created at higher confinement strength
values are also potentially indicative of higher Reynolds number effects. At
higher Reynolds numbers the turbulent boundary layer is a thin sheet of vortices
and appears similar to the images in Figure 24 and Figure 25.

68

Figure 24. Velocity magnitude contour for immersed plate with VC2 and confinement
strength of 0.004. The boundary layer is exhibiting roll-up behavior.

Figure 25. Vorticity magnitude contour for immersed plate with VC2 and confinement
strength of 0.004. The boundary layer is exhibiting roll-up behavior.

69

Figure 26. Velocity magnitude contour for immersed plate with VC2 and first-order upwind
differencing for the convection term. The added numerical diffusion causes the boundary
layer to thicken and suppresses the transient instabilities.

Figure 27. Velocity magnitude contour for immersed plate with VC2 and second-order
upwind differencing for the convection term, but with a reduced confinement strength of
0.0035. Boundary layer instabilities are no longer present in the solution.

70
As shown in previous cases, the balance of confinement strength with
numerical diffusion provides for varying effects on the flow. This can be used to
an advantage to study Reynolds number effects on the solution; at higher
confinement strength values, the flow becomes more chaotic which is indicative
of an effective higher Reynolds number. Smaller confinement strength values
result in a lower effective Reynolds number, and removing Vorticity Confinement
altogether results in a laminar Navier-Stokes solution. These effects will be
studied more thoroughly in 5.4.
5.3 Advection of a Compact Vortex
The previous two case studies have used a steady-state solver to solve
boundary layer problems and provide an initial estimate for the confinement
strength parameter. The true applicability of Vorticity Confinement to influence
vortical regions, however, is through enhancement of transient aspects of the
flow and thus the implicit time-stepping time-accurate solver pisoVcFoam has
been developed for the remaining case studies. One of the characteristics of VC
is the ability to amplify and sustain vortical features by treating them as solitary
waves which ride on the convecting fluid. The solitary wave concept is similar to
the soliton approach used by condensed matter physicists, in which dispersive
effects spreading a wave-packet are balanced by self-steepening terms creating
a stable configuration [68]. Locally, the vortical feature modified by VC behaves
as the full asymptotic solution, whereas conventional methods supply the
intermediate asymptotic solution which eventually asymptotes to a constant value
of zero. VC effectively creates a non-zero constant asymptote in vortical regions.
To demonstrate this behavior, a single compact vortex is advected through a
periodic flow domain with a fixed background velocity.
5.3.1 Setup Details
The advection of the compact vortex is presented in the domain pictured
in Figure 28. A uniform 401 x 201 cell mesh is used to discretize the two-

71
dimensional domain which is decomposed for parallel computation over 8
processors. Domain decomposition is performed by the METIS library which is a
decomposition method that minimizes the number of processor boundaries. The
odd number of cells in each coordinate direction is necessary for the vortex
center to be defined at exactly (0, 0) within a single cell; otherwise, four equal
point vortices are created around the (0, 0) point which behave as four compact
vortices instead of a single vortex. The compact vortex is imposed on top of a
background velocity of 0.1 m/s which convects the vortex through the domain.
The domain is bounded by symmetry planes in the spanwise direction and
periodic boundaries in the streamwise direction which allow the vortex to pass
from one side of the domain to the other during the calculation. Explicit diffusion
of = 0.001 m
2
/s is added for the VC model. Each time-accurate simulation is
run for 40 seconds with a streamwise CFL number of 0.1; based on the
background velocity this should result in a full transit of the vortex through the
domain (4 m). Three cases are studied with explicit numerical diffusion of 0.001
m
2
/s: a baseline case without VC, and two cases with the VC1 and VC2
formulations, respectively. Due to the low background velocity relative to the
vortex strength, the vortex simulations provide an opportunity to examine the
conservative aspects of both the VC1 and VC2 formulations.

Figure 28. Boundary conditions for compact vortex advection case study.

72
5.3.2 Numerical Stability Requirements
This case study is the first time-accurate simulation examined in this work
and provides an opportunity to investigate the stability of the Vorticity
Confinement algorithm with a time-accurate solver and determine whether the
calibrated confinement strength coefficients from the steady-state simulations are
applicable. The gradient discretization schemes used in the steady-state
simulations for gradient calculations were found to be unstable during initial tests
of the vortex simulation. To ensure stability, cell limiting was enabled for the
gradient computation in all VC terms as well as the convection term. Cell limiting
limits the extrapolated face values for the gradient computation based on the
maximum and minimum cell and neighbor values, which helps prevent numerical
instability by adding a slight amount of local dissipation. Using cell limiting with
the VC1 scheme proved to be sufficient to maintain the highest calibrated
confinement strength of 0.25 without observing any oscillations in the solution.
For the VC2 confinement term harmonic differencing (additional to
harmonic mean) is also necessary for stability during time-accurate simulation.
This differencing scheme interpolates the inverted vorticity field from the cell
centers to the faces for gradient computation and is weighted by the magnitude
of vorticity at each cell, minimizing the gradient calculation when neighboring
values are small. The operation performs a stabilizing function similar to the
method suggested by Steinhoff and Lynn [7] in which VC2 terms exhibiting
alternating signs with neighboring values are set to zero. With the harmonic
differencing scheme the VC2 confinement strength was also reduced to 0.003
(3) to obtain a stable vortex within the calculation. At higher confinement
strength values the flow becomes perturbed near the vortex and results in a fully
chaotic flow field with large scale vortices appearing in the solution. This behavior
is not unexpected; due to its conservative properties the VC2 formulation
conserves angular momentum and generates higher velocities near the center of
the vortex with higher confinement strength values. Higher velocities tend to
cause shearing of the axisymmetric vortex structure. The effect becomes strong

73
enough at the highest confinement strength to cause vortex breakdown, which
may also be indicative of higher Reynolds number flow behavior [69].
5.3.3 Simulation Results
The streamwise vorticity magnitude across the centerline of the domain is
presented in a one-dimensional plot for all three vortex cases at 1 s duration (100
time steps) in Figure 29. Due to the lack of a spanwise background velocity
component the vortex does not deviate from the centerline of the domain during
the calculation. In the baseline case the vortex dissipates and spreads as it is
advected by the background flow; the vorticity magnitude has decreased one
order of magnitude after only one second. Using either the VC1 or VC2
formulation, however, results in a much higher vorticity magnitude with the vortex
focused around a region 2-3 grid cells thick. Figure 30 displays the
corresponding vorticity magnitude contour plot at one second duration. The scale
for the baseline case is reduced to highlight the spreading of the vortex.

Figure 29. Centerline vorticity magnitude after 1 s (100 time steps). Without VC the vortex
strength has decreased an order of magnitude.

74

Figure 30. Contour plots of vorticity magnitude at one second simulation duration (100
time steps). Note reduction of scale for simulation without VC.

75
Figure 31 displays the streamwise centerline vorticity plot for all three
cases at 10 seconds duration (1000 time steps). In the baseline case the vortex
has effectively disappeared within the background flow, while both VC1 and VC2
prevent the vortex from spreading beyond a 2-3 grid cell region spanning the
centroid of the vortex. By 10 seconds it has also become evident that the VC1
solution has begun to lag both the VC2 and no-VC solutions. Steinhoff et al. [15]
have previous shown in analytical form that VC1 conserves total amplitude but
not the position of the centroid, while VC2 conserves the position of the centroid.
As a result, the position of a vortex will deviate from the correct value based on
the original governing equations when using the VC1 formulation over a long
period of time in a slow moving background flow. The VC2 formulation, however,
is able to maintain the position of the centroid and as a result does not exhibit the
lagging behavior. Figure 32 displays the contour plots of vorticity magnitude at
the 10 second duration corresponding to Figure 31.

Figure 31. Centerline vorticity magnitude after 10 s (1000 time steps).

76
Besides the change in position, the vorticity magnitude contours for both
cases are not significantly different from the previous contours at one second in
Figure 30. The VC model effectively behaves as a non-zero full asymptotic
solution within the local region of the vortex and does not allow further
dissipation. The contour for the baseline case, however, is not even shown
because the vortex has dissipated fully such that it essentially blends with the
background flow. Without VC the intermediate asymptotic solution of the vortex
tends towards the asymptotic solution of zero vortical flow within the constant
background solution.
The streamwise centerline vorticity plot at 40 seconds duration (4000 time
steps) is shown in Figure 33 along with the corresponding vorticity magnitude
contours in Figure 34. Based on the background velocity the vortex should make
a full transit through the periodic boundary back to the initial starting point after
40 seconds.

Figure 32. Vorticity magnitude contour after 10 s (1000 time steps). Baseline results not
shown because the vortex has been absorbed into the background flow.

77

Figure 33. Centerline vorticity magnitude after 40 s (4000 time steps).

Figure 34. Vorticity magnitude contour after 40 s (4000 time steps). Low magnitude
vorticity concentrations are visible in VC1 solution.

78
As expected, the VC1 solution deviates considerably from the unperturbed
location of the vortex centroid, achieving only approximately 80% of the
necessary travel distance. The VC2 solution, on the other hand, achieves
approximately 97.5% of the anticipated travel distance. Low-magnitude
asymmetric oscillations are evident in the VC2 solution near the vortex centroid;
it is surmised that over the 40 second duration the asymmetries are sufficient to
perturb the centroid position by the marginal amount. Low magnitude vorticity
concentrations are evident in the VC1 contour plot in Figure 34 at 40 seconds
which were not evident in previous plots. These low magnitude secondary
vortices are numerical artifacts which result from small amounts of vorticity
generated from the impulsive start of the initial condition. Over the full duration of
the calculation these pockets of diffused vorticity are compressed to form
secondary vortical regions.
To complete the full 40 second simulation on 8 processors required a total
run time of 2831 seconds for the baseline case, 3195 seconds for the VC1 case,
and 3202 seconds for the VC2 case. The time required for the VC routine does
not account for the 13% increase in required simulation time. The additional time
results from the solution of the Poisson equation solver for the pressure-
correction step, which is the main driving component for solution speed in the
PISO and SIMPLE algorithms. The pressure correction step is solved in the
vortex simulation using a Conjugate Gradient Krylov subspace solver with a
diagonal incomplete-Cholesky preconditioner. In this case, the inclusion of VC
appears to cause increased sub-iterations of the Conjugate Gradient solver,
suggesting that a different preconditioner could be more efficient for the
simulation. In other situations VC may assist the preconditioner and drive the
solution faster towards convergence. This behavior is similar to the effects of
other numerical methods and cannot be predicted but must be evaluated on a
case-by-case basis to determine the optimal combination of solution algorithms.

79
5.4 Flow over a Backward-facing Step
The previous studies have demonstrated the performance of the
developed Vorticity Confinement solvers to approximate thin turbulent boundary
layers and convect vortices over long distances. These simple simulations,
although educational regarding the observed model characteristics, are only two-
dimensional whereas true turbulent flow is three-dimensional. A classical three-
dimensional problem studied numerically and experimentally is the backward-
facing step or rearward-facing step, a simple geometry which nonetheless
induces a variety of flow regimes including boundary layers, a mixing layer,
reattachment and flow reversal. Due to the complexity of the generated flow, the
backward-facing step has proven to be an excellent test case for the validation of
turbulence models, in particular large-eddy simulation (LES) models.
Experimental studies conducted by Pitz and Daily [70] and Keller et al. [71] with
non-reacting and reacting flows using a standard backward step geometry have
been replicated numerically for the purpose of validating LES combustion models
[72,73]. For this case study, experimental and numerical results from the non-
reacting studies will be compared against time-averaged data from studies
utilizing the VC1 and VC2 formulations. These data include mean velocity and
RMS velocity profiles at consecutive vertical spans downstream of the step, as
well as reattachment lengths. RMS velocity indicates the level of turbulent
fluctuations present in the solution, while reattachment length represents the
distance to the point of boundary reattachment for the shear layer. Reattachment
length has been found in previous studies to be a function of Reynolds number
as well as the separating boundary layer characteristics [71]; as a result, it is
expected that varying the confinement strength parameter will have an effect on
reattachment length. To maintain consistency with prior studies for comparison,
reattachment length is defined as the extent of the reverse streamwise velocity in
the circulation zone.

80
5.4.1 Setup Details
The backward-facing step geometry replicating the experimental setup of
Pitz and Daily [70] is shown in Figure 35. The streamwise (x), normal (y), and
spanwise (z) dimensions have been normalized by the step height h = 0.0254 m.
The experimental Reynolds number Re = 22,100 is based on the step height and
an average inlet velocity U = 13.3 m/s. The domain extends for approximately 8
step lengths before narrowing in a smooth contraction to the exit plane. This
geometry is identical to the domain used for the LES study performed by Park
and Ko [73]. Figure 36 highlights a representation of the mesh, a non-uniform
grid comprised of 248 x 57 x 20 cells in the streamwise, normal, and spanwise
directions, which is similar to that used in the Park and Ko LES study. Wall
functions are used for the boundary layer in the LES study, and as a result the
mesh is not overly refined near the boundaries. This non-uniform LES-typical
mesh is used for the initial Vorticity Confinement studies to ensure an accurate
comparison to both experimental data as well as the numerical studies.
To account for the variation in cell size, the confinement strength is scaled by a
non-dimensional proportion of local cell volume to the average cell volume for the
mesh, similar in concept to the scaling method proposed by Butsuntorn [33]. Wall
boundaries are set to a no-slip condition while the spanwise-normal surfaces are

Figure 35. Schematic representation of backward-facing step geometry (h = 0.0254 m).

81
specified as periodic boundaries to simulate the larger spanwise distance
between walls in the experiment.
As stated previously, an implicit second-order backward differencing
temporal scheme is used instead of the implicit Crank Nicholson scheme to
match the setup of the LES simulation. Each case is run for an initial duration of
0.1 seconds which allows for approximately four flow exchanges through the
domain to establish a regular flow pattern. Statistical quantities are subsequently
time-averaged over an additional 0.1 seconds with a time step of 1E-05 s,
equivalent to a maximum CFL number of approximately 0.2, to ensure a
sufficient statistical dataset.
5.4.2 Mean Flow Comparison
The case studies presented here are compared to the experimental
results of Pitz and Daily [70] and the numerical LES results of Park and Ko [73] at
profiles of streamwise mean velocity and streamwise RMS velocity defined at
specific locations downstream of the step, as well as measured and computed
reattachment lengths. The LES simulation is chosen as an example of accurate
conventional turbulence modeling using a similar methodology to VC. LES
resolves the large scales of turbulence directly using cell size as an explicit scale
filter and computes interactions with the smaller scales through an explicit sub-
grid model [74]. The sub-grid model acts to dissipate the under-resolved smaller
scales. In comparison, an implicit large-eddy simulation (ILES) approach requires

Figure 36. LES-typical non-uniform mesh used for initial VC studies in 5.4. Image depicts
the 248 x 57 cell mesh in the X and Y coordinate directions; 20 uniform grid cells are used
in the Z coordinate direction (normal to plane).

82
no sub-grid model, but instead relies on the inherent numerical dissipation to
provide proper turbulence decay. Vorticity Confinement can be generalized as a
type of ILES method which resolves scales corresponding to the cell size and
models the effects of the sub-grid scales through capture of boundary layers and
vortex sheets, assumed to be the smallest scales [15]. With proper resolution
and confinement parameters VC should result in a similar accurate solution to
the explicit LES simulation.
To provide a more direct comparison with the VC models, an additional
case is solved on the mesh in Figure 36 with VC switched off resulting in a
laminar Navier Stokes solution on the backward-facing step geometry with no
explicit dissipation other than that provided by the physical viscosity necessary to
achieve the appropriate Reynolds number given in the experiment. The case
provides a baseline solution to contrast with the VC results. Differences highlight
the characteristics of the flow which are truly affected by the VC model while
similarities establish which base flow phenomena are resolved due to the level of
grid refinement.
A list of the studied cases detailing simulation parameters is given in Table
1 and includes the computed reattachment lengths based on the time-averaged
streamwise velocity. For the backward-facing step the stability of the VC1 model
Table 1. Simulation parameters for initial simulations in 5.4.

Methodology value value Reattachment length X/h* (accuracy 0.5h)
No VC 0.0 1.536E-5 5.8
VC1 0.25 1.536E-5 6.5
VC2 0.0016 0.001 5.8
VC2 0.0020 0.001 5.5
VC2 0.0025 0.001 4.4
*Reattachment length computed from time averaged streamwise velocity varies over time
with stated accuracy. LES and experimental studies report reattachment lengths of 6.8 and
7.0, respectively, with similar accuracy.

83
was found to be invariant with respect to the value of and as a result was set
equal to the laminar viscosity. Since the parameter represents both physical
and numerical viscosity this suggests that the implicit numerical diffusion inherent
in the discretization scheme was sufficient to balance the confinement parameter
without any added explicit diffusion. The behavior of the VC2 scheme, however,
was found to be heavily dependent on the ratio of to . The value of was held
at a constant 0.001 m
2
/s while various confinement values between 1.5 and
3.0 were investigated. The VC2 cases mentioned here represent the most
appropriate range for discussion matching the Pitz and Daily experimental data.
As stated in Table 1, the experimental and LES studies report
reattachment lengths of 7.0 and 6.8, respectively, with an accuracy of 0.5h.
Reattachment length grows with increasing laminar Reynolds number, but
reduces to a constant value dependent on geometry after transition to turbulent
Reynolds number [75]. For the studies here, reattachment length is measured by
observing the extent of the recirculation zone defined by the mean streamwise
velocity. A representative iso-volume plot is shown in Figure 37 for the laminar
solution; iso-volume plots for the remaining cases are found in the Appendix in
Figure A-1 through Figure A-4 for reference. The iso-volume gives a volumetric
representation of all values of mean streamwise velocity less than zero. While
the reattachment point for the VC1 case lies within the tolerance band defined in
the experiment, the VC2 cases exhibit a reduction in reattachment length as is
increased, deviating from the constant experimental value around 7.0. Variation
of the reattachment point has been studied by Nait Bouda et al. [76] who report
that the reattachment point varies in time about its mean position and is affected
by the large eddies generated within the shear layer. Furthermore, their research
indicates that larger eddies can cause the reattachment point to fluctuate and
cause the mean reattachment length to decrease. The reduction in reattachment
length for the higher VC2 cases suggests that the VC2 model is enhancing the
large eddies in the shear layer enough to cause a reduction in mean
reattachment length, similar to the effect reported by Nait Bouda et al [76].

84
The time-averaged streamwise velocity contour for the laminar solution
(no VC) in Figure 38 is compared with a streamwise contour for the VC1 case in
Figure 39. The VC1 solution exhibits the most significant difference in the
confinement of the boundary layer of the upper wall near the beginning of the
contraction; in the laminar solution the boundary layer thickens at this point which
accompanies a shorter recirculation zone directly below at the lower wall. The
time-averaged flow characteristics of the VC1 case compared to experimental
data, LES results, and the laminar solution are given in Figure 40 and Figure 41.
The mean flow profiles for the laminar solution and the VC1 solution in Figure 40
compare well with both experimental data and the LES solution. The VC1 model
shows small regions of improvement in mean velocity profile at X/h = 1 and X/h =
3 within the recirculation region (-1 < Y/h < 0) compared to the laminar solution
which suggest that the VC1 model more accurately captures the circulatory
behavior. The streamwise RMS velocity profiles for the laminar solution and VC1
solution in Figure 41 show satisfactory comparison to the experimental data.

Figure 37. Iso-volume of mean streamwise velocity showing region of U < 0.0 for the
laminar solution (no VC).

85

Figure 38. Mean streamwise velocity contour for laminar solution (no VC).

Figure 39. Mean streamwise velocity contour for VC1 solution. Contour levels from Figure 38 apply.

86

Figure 40. Mean streamwise velocity profiles from VC1 solution compared to experimental
data, LES results, and the laminar solution. Y-axis position has been normalized by step
height, while velocity has been normalized by the prescribed inlet velocity.

87

Figure 41. Streamwise RMS velocity profiles from the VC1 solution compared to
experimental data, LES results, and the laminar solution.

88
On average the laminar solution under-predicts the level of turbulent
fluctuations in the free stream, seen as a deficit in the RMS velocity plot at X/h =
5 compared to the experiment profile. Further deviations of the laminar solution
from experimental data are seen as an over-prediction of RMS velocities in the
region near the boundary layer at all profiles downstream of X/h = 1. In contrast
to the laminar solution, the VC1 model appears to more accurately predict the
RMS velocity profile, as indicated by a close correspondence to the shape of the
experimental data in the free stream. In particular the result at X/h = 1 matches
the peak experimental velocities within the shear layer. Due to the application of
Surface Confinement the freestream velocity is maintained to a point very close
to the wall, resulting in an over-prediction of boundary layer velocities near the
bottom wall at X/h = 5 and X/h = 7. This behavior is expected as VC does not
allow the boundary layer to diffuse. In this VC1 study both Field Confinement and
Surface Confinement use a confinement value of = 0.25; the over-prediction at
the boundary layer indicates better results may be achieved near the boundary
by varying the confinement strength for Surface Confinement.
Mean streamwise velocity contours for the VC2 solutions with varying
confinement strength values are presented in Figure 42. It becomes apparent
from the comparison between cases that at lower values the boundary layer is
still allowed to thicken, similar to the behavior seen in the 5.1 case study. An
increase in the value of results in a domain-wide decrease in boundary layer
thickness, as well as an increase in velocity along the top wall. The behavior is
similar to an increase in Reynolds number, in the sense that VC2 at higher
confinement strengths counteracts more numerical diffusion within the solution,
resulting in a lower effective flow viscosity (and therefore higher effective
Reynolds number). Due to this effect, there can be a fine line between too little
confinement and too much confinement with VC2 when comparing to
experimental data obtained at a certain Reynolds number.

89

Figure 42. Mean streamwise velocity contours for VC2 solutions; = 1.6 (top), = 2.0 (middle), = 2.5 (bottom).

90
The time-averaged flow characteristics of the VC2 cases compared to
experimental data, LES results, and the laminar solution are presented in Figure
43 through Figure 48. The mean velocity profiles from the = 1.6 solution in
Figure 43 do not correlate with the experimental data any better than the laminar
solution, with the exception of the X/h = 7 profile in which the VC2 solution lies
directly on the experimental profile. The RMS velocity profiles in Figure 44 at this
confinement strength, however, provide the closest match to the experimental
data by resolving both the peak RMS magnitudes as well as the overall profile
shapes at each measurement location. Realistically, the underlying laminar
solution already provides satisfactory resolution of the mean velocity profile; the
marginal level of confinement is enough to affect the RMS fluctuations but not
provide any improvement in the mean velocity profiles. The time-averaged
velocity profiles for = 2.0 in Figure 45 show some improvement with the higher
confinement strength at X/h = 1, where the recirculation velocity has shifted to
correspond to the given data. The shapes of the mean velocity profiles at both
X/h = 1 and X/h = 3 provide a better correspondence to the experimental data
within the shear layer than even the reference LES solution. However, in Figure
46 the RMS velocity profiles begin to show evidence of over-prediction within the
recirculation region and near the boundaries. This indicates that the level of RMS
fluctuations has exceeded what was measured during the experiment. Increasing
the confinement strength further to = 2.5 leads to excessive overshoot near
the boundaries in both the mean velocity profiles in Figure 47 and the RMS
velocity profiles in Figure 48. The over-prediction of the RMS velocity is present
in only the vortical regions of the flow, as evidenced by the RMS profile at X/h = 1
in which the portion of the profile crossing the inlet stream matches the
experimental measurements, while the lower portion of the profile within the
recirculation zone exhibits more than 2x increase in RMS magnitudes. The
behavior suggests that for this type of internal flow where the boundary layer
thickness is important a more conservative value of the confinement strength
may be necessary to provide a reasonable correlation to experimental data.

91

Figure 43. Mean streamwise velocity profiles from the VC2 = 1.6 solution compared to

92

Figure 44. Streamwise RMS velocity profiles from the VC2 = 1.6 solution compared to

93


94


95


96


97
5.4.3 Instantaneous Flow Comparison
Time-averaged experimental data comparisons provide a good method to
gauge the accuracy of simulation results; however, to gauge the potential of a
particular model instantaneous flow field comparisons can also be an effective
tool. Figure 49 displays horizontal section slices of the downstream vorticity flow
field at the Y = 0 height (level with the step height) for each case at the 0.2 s time
step. The plots effectively provide a cross-section of the shear layer for each
case, allowing direct comparison of the baseline laminar solution with the VC1
and three VC2 solutions. The flow is characterized by the separation at the
leading edge of the step; the shed vorticity sheet rolls up into a three-dimensional
vortical structure which initially spans the domain and then breaks up into smaller
vortices. This is observed in Figure 49 as an initial continuous sheet of high
vorticity near the step which breaks up into individual sheets and vortices with
lower vorticity. The laminar solution, VC1 solution, and VC2 = 1.6 solution
show similar behavior of the initial vorticity sheet which is supported by the
previous correlations with experimental data. The VC2 = 2.0 and = 2.5
solutions exhibit much earlier break-up of the initial vorticity sheet which explains
the poor correlation in the shear layer at X/h = 1 in Figure 47; at this station the
shear layer has already begun to mix and shows less separation compared to the
experiment. The higher VC2 simulations overall show evidence of a more
chaotic, robust flow field downstream compared to the laminar solution.
To provide an encompassing picture of the temporal aspects of the flow
field, flow visualizations of vorticity for each case spanning 0.1 second duration
are included as attachments. Each case includes a vertical center plane
visualization of vorticity magnitude and a three-dimensional volumetric-rendered
visualization of vorticity. Instantaneous images which correspond to the flow
visualizations are displayed in the text for reference. Figure 50 displays center
plane vorticity slices from VISUALIZATION 1 and VISUALIZATION 3 for the
laminar and VC1 solutions, while Figure 51 displays three-dimensional vorticity
images from VISUALIZATION 2 and VISUALIZATION 4 for the same cases.

98

Figure 49. Horizontal section slices at Y = 0 height, level with the step leading edge.

99

Figure 50. Instantaneous center plane slices of vorticity magnitude, from VISUALIZATION 1, laminar solution (top) and
VISUALIZATION 3, VC1 solution (bottom).

100

Figure 51. Instantaneous volumetric-rendered images of vorticity magnitude, from
VISUALIZATION 2, laminar solution (top) and VISUALIZATION 4, VC1 solution (bottom).
Contour levels below 2000 are invisible to accentuate large coherent structures. Upper
wall solution has been clipped to emphasize shear layer structures.

101
The laminar solution exhibits fewer vortical structures which begin to
dissipate by approximately X = 6h. The VC1 solution, on the other hand, shows
an enhancement of the vortical structures from the beginning of the shear layer
continuing past the beginning of the contraction. VC1 appears to magnify the
vortical structures compared to the laminar solution and the effect of surface
confinement with VC1 prevents the thickening of the boundary layer across the
top wall which is evident in the laminar solution.
Similar instantaneous images of vorticity are presented from
VISUALIZATION 5 through VISUALIZATION 10 for the three VC2 solutions in
Figure 52 and Figure 53. Increasing the confinement parameter clearly has an
enhancing effect on the vortical structures, causing the shear layer to mix sooner
downstream of the step and finer vortical structures to appear at the highest
values. Particularly evident in VISUALIZATION 7 and VISUALIZATION 9 is
boundary layer separation from the top wall induced by nearby vortical structures
in the flow field. As stated in the 5.1 case study discussion, the boundary layer is
allowed to separate under the influence of an adverse pressure gradient but can
still reattach downstream due to the effects of the VC model. Another notable
effect seen in VISUALIZATION 10 is the appearance of random temporary gaps
in the flow field. This interesting result demonstrates that the VC model does not
prescribe the behavior of the vortical structures, but simply enhances the
structures as they propagate downstream, preserving the random and chaotic
nature of the turbulent flow field.
In the = 2.0 and = 2.5 solutions both large and small scale structures
are accentuated within the shear layer; the VC2 model prevents the structures
from dissipating through the exit of the domain, causing the boundary layer to
reattach sooner in the high VC2 cases. It is possible that continued
enhancement downstream is a result of the non-uniform cell correction; within the
contraction larger cells are present which create larger corrections to account for
the cell size variation. As a result, the VC model may be causing the downstream
flow in this region to recover too quickly compared to the experiment.

102

Figure 52. Instantaneous center plane slices of vorticity magnitude, from VISUALIZATION 5 with VC2 = 1.6 (top),
VISUALIZATION 7 with VC2 = 2.0 (middle), and VISUALIZATION 9 with VC2 = 2.5 (bottom).

103

Figure 53. Instantaneous volumetric-rendered images of vorticity magnitude, from VISUALIZATION 2, baseline laminar solution
(top left), VISUALIZATION 6, VC2 = 1.6 (top right), VISUALIZATION 8, VC2 = 2.0 (bottom left), and VISUALIZATION 10, VC2
= 2.5 (bottom right). Contour levels below 2000 are hidden to accentuate large coherent structures. Upper wall solution has
been clipped to emphasize shear layer structures.

104
5.4.4 Turbulence Kinetic Energy Spectra
To provide qualification of the turbulent solution, a turbulence kinetic
energy (TKE) spectrum versus wavenumber can be generated from the velocity
flow field. Real turbulent flows are characterized by the transfer of energy
continuously from the largest scales governed by the mean flow to the smallest
molecular scales, where energy is dissipated as heat on the molecular level. This
energy cascade from large to small scales was observed by Kolmogorov to
exhibit similarity at the smallest scales for turbulent Reynolds numbers [77]. The
Kolmogorov similarity hypotheses state that for sufficiently high turbulent
Reynolds number there is an eddy range known as the inertial sub-range which
falls between largest and smallest eddies. Within this range, the turbulent
motions are statistically isotropic and the energy cascade process is independent
of both mean flow and viscous effects, resulting in a universal form with a
transfer rate equal to -5/3 logarithmic slope [77]. Extending this theory to
theoretical and numerical calculation, a turbulent solution must exhibit a -5/3
slope region in some portion of the inertial sub-range, otherwise the solution
suffers from improper turbulence energy cascading which may result from
incorrect turbulence modeling [78]. The TKE spectrum as a function of
wavenumber is generated by a Fourier transform of temporal data from a single
point by using Taylors frozen turbulence approximation, which assumes that
spatial correlations (wavenumber) can be approximated by temporal correlations
(frequency). In regions of regularly periodic turbulence it provides a necessary
metric for turbulence quantification [77].
The non-dimensional TKE spectra as a function of non-dimensional
wavenumber are presented for the VC2 = 1.6 solution in Figure 54 and =
2.5 solution in Figure 55. Additional TKE spectra for the VC1 case and the VC2
= 2.0 solution can be found for reference in the Appendix in Figure A-5 and
Figure A-6. Each plot presents the respective TKE spectrum along with a
spectrum generated from the laminar solution at the same measurement
location; the -5/3 slope has been included for reference. The measurement

105
location has been chosen at approximately X/h = 2 to ensure the presence of
periodic turbulence for the correlation. The spectrum and wavenumber have
been non-dimensionalized by the reference step height h, the mean velocity over
the measurement period, and the mean TKE magnitude over the measurement
period. Both VC2 spectra in Figure 54 and Figure 55 exhibit the -5/3 slope in the
approximate center of the wavenumber range, consistent with the power-law
spectrum relationship expected by the Kolmogorov hypotheses. The laminar
solution also exhibits a region of -5/3 slope; however, this covers a narrower
range of wavenumbers and quickly begins to steepen in slope at relatively low
wavenumber. Steepening of the slope occurs as the wavenumber enters the
dissipation range. Faster falloff of the -5/3 slope does not necessarily indicate
that the solution is incorrect; however, it can be a precursor to poor resolution
[79]. For all cases the TKE spectra asymptote in the dissipation range to a non-
similar slope compared to the theoretical Kolmogorov TKE spectrum. This
flattening of the dissipation slope is a behavior shared with ILES-type methods;
at best only the inertial range for an ILES fluid is expected to be similar to the
behavior of a real fluid and thus the non-similarity in the dissipation range can be
disregarded [80].
Another observation regarding the TKE spectra of the VC2 solutions is the
expansion of the characteristic -5/3 slope inertial wavenumber region. Energy
cascading begins at lower wavenumbers compared to the laminar solution;
furthermore, as is increased the -5/3 slope region extends further to higher
wavenumbers when comparing the = 1.6 and = 2.5 spectra. The behavior is
again similar to that of an ILES fluid, such that an enlargement of the -5/3 slope
inertial range is indicative of higher effective Reynolds number flow [80]. This
suggests that the turbulent structures observed in the visualizations at higher
values are truly the result of an effective higher Reynolds number. Extension of
the inertial range to higher wavenumbers also suggests that the VC model is
resolving subscale structures not captured in the laminar solution.

106

Figure 54. TKE spectrum versus wavenumber for VC2 = 1.6 solution.

107

Figure 55. TKE spectrum versus wavenumber for VC2 = 2.5 solution.

108
5.4.5 Coarsened Mesh Solutions
One of the potential effects of the VC2 model resolving more turbulent
structures is the relaxation of mesh density requirements while still providing a
reasonably accurate solution. To investigate this possibility, the backward-facing
step mesh was coarsened in the X and Y coordinate directions. The subsequent
mesh is shown in Figure 56 and represents a relatively uniform mesh of 127 x 35
x 20, for a total cell count of approximately 85,000 cells. The refinement in the Z
direction was left identical to the original mesh with 20 spanwise grid cells in
order to preserve reasonably isotropic cell shapes. While the larger mean cell
size will allow a larger time step without exceeding CFL criteria, the same time
step was used to ensure conformity of statistical data to the previous results.
Initial studies with an identical parameter from previous cases indicated
excessive dissipation within the solution; the coarser mesh inherently generates
higher numerical dissipation. As a result, for the coarse mesh cases is halved
to 0.0005 and confinement strength values of = 2.0 and = 2.5 are chosen
for comparison to a baseline laminar solution.
The mean streamwise velocity profiles for the coarse mesh solutions are
depicted in Figure 57. The laminar solution shows slightly more boundary layer
separation at the top wall compared to the fine mesh solution in Figure 38,
indicative of the higher dissipation present on the coarse mesh. The two VC2
solutions, however, are dramatically different from the VC2 solutions on the fine

Figure 56. Coarsened backward-facing step mesh. 127 x 35 x 20

109

Figure 57. Mean streamwise velocity contours for coarse mesh solutions; baseline laminar solution (top), VC2 = 2.0 (middle),
VC2 = 2.5 (bottom).

110
mesh in Figure 42 with identical confinement strength. The previous solutions
tend to over-predict the velocity, whereas the coarse mesh solutions now show a
good comparison with the baseline laminar fine mesh contour. This difference
highlights the effect of the tuned balance between and ; achieving the correct
balance between the two parameters leads to a realistic result.
The time-averaged characteristics of the coarse mesh cases are portrayed in
Figure 58 and Figure 59. The LES comparison has been omitted since the mesh
is no longer similar. With mean velocity profiles both VC cases perform well in
the shear layer at X/h = 1, with the = 2.5 case exhibiting a better prediction of
the mean reverse velocity magnitude in the circulation zone than the laminar
solution. At X/h = 3 all the coarse solutions appear to recover in the shear layer
too quickly compared to the experiment; however, the VC solutions still predict
the correct reverse velocity magnitude. Further downstream it is clear that in the
VC cases the flow reattaches to the bottom wall sooner than the experiment
suggests, causing a deviation of the mean flow profile. The RMS velocity profiles
in Figure 59 indicate a much closer match between the VC solutions and the
experimental data with regards to turbulent fluctuations; the laminar solution in
general under-predicts the RMS fluctuations and does not even match the profile
shape at the X/h = 5 and X/h = 7 stations, while both VC solutions show excellent
agreement with the experimental profiles at all four stations.
Flow visualizations of vorticity are also generated for the coarse mesh
cases to explore the temporal variations of the flow. Instantaneous center plane
vorticity magnitude contours and volumetric-rendered contours are shown from
respective visualizations in Figure 60 and Figure 61 for the baseline laminar
solution and two VC2 solutions on the coarse mesh. The enhancement of
turbulent vortical structures within the flow as the value of is increased is
particularly evident compared to the baseline laminar solution. The shed vorticity
sheet at the beginning of the shear layer in the laminar solution extends almost
two step lengths before beginning to roll up and separate. The VC2 cases, on

111

Figure 58. Mean streamwise velocity profiles from the coarse mesh comparing
experimental data, the coarse laminar solution, the VC2 = 2.0 solution, and the VC2 =
2.5 solution.

112

Figure 59. Streamwise RMS velocity profiles from the coarse mesh comparing
experimental data, the coarse laminar solution, the VC2 = 2.0 solution, and the VC2 =
2.5 solution.

113

Figure 60. Instantaneous center plane slices of vorticity magnitude on the coarse mesh, from VISUALIZATION 11 with the
baseline laminar solution (top), VISUALIZATION 13 with VC2 = 2.0 (middle), and VISUALIZATION 15 with VC2 = 2.5
(bottom).

114

Figure 61. Instantaneous volumetric-rendered images of vorticity magnitude on the coarse mesh, from VISUALIZATION 12,
laminar solution (top left), VISUALIZATION 14, with VC2 = 2.0 (top right), and VISUALIZATION 16, with VC2 = 2.5 (bottom).
Contour levels below 2000 are hidden to accentuate large coherent structures. Upper wall solution is clipped as in Figure 53.

115
the other hand, show progressively quicker roll-up behavior of the vortical sheet
in the shear layer and exhibit more turbulent structures within the flow. The
boundary layer on the coarse mesh VC2 cases appears thicker than the
boundary layer on the corresponding fine mesh cases due to the coarsened grid.
Vortex-induced separation of the boundary layer is particularly evident in
VISUALIZATION 13 and VISUALIZATION 15. In comparison to the behavior
exhibited on the fine mesh, the behavior of the VC model on the coarser mesh is
more subdued. The inherent numerical dissipation dominates the solution and
reduces the effect of the VC confinement parameters over the same range.
Additionally, the structures in the coarse mesh VC2 cases are much larger and
more coherent than the finer chaotic structures created in the high fine mesh
VC2 cases observed in Figure 53. The coarser mesh acts as an implicit filter
such that mesh size governs the highest resolvable wavenumber cutoff scale
above which turbulence cascading is governed by numerical diffusion. As a
result, the smallest turbulent structures resolved on the coarser mesh will be
larger than those resolved on the finer mesh. Nonetheless, both coarse and finer
mesh cases show a good correlation to mean flow data indicating that the
important flow features are captured in both cases.
The corresponding TKE spectra are displayed in Figure 62 for the VC2 =
2.0 case and Figure 63 for the VC2 = 2.5 case. The laminar TKE spectrum
exhibits a very narrow wavenumber range corresponding to the -5/3 slope. In
comparison, both coarse mesh VC solutions reveal a central range of
wavenumbers at the -5/3 slope, and increasing results in an expansion of the -
5/3 slope region similar to the behavior of the fine mesh cases. Compared with
the fine mesh spectra, the coarse mesh spectra display a flattening of the slope
in the dissipation range beginning at a lower wavenumber. ILES simulations
exhibit a wavenumber range with steep decay followed by a flattening of the
slope near the grid-scale wavenumber [80]. The trend of flattening of the slope at
a lower wavenumber with a coarser grid appears consistent with this finding and
confirms further the similarities of VC with ILES methods.

116

Figure 62. TKE spectrum versus wavenumber for coarse mesh VC2 = 2.0 solution.

117

Figure 63. TKE spectrum versus wavenumber for coarse mesh VC2 = 2.5 solution.

118
5.5 Complex Immersed Body Simulation
Many flows of engineering interest require the resolution of complex
physical features while simultaneously solving complicated flow fields at turbulent
Reynolds number; to achieve accurate solutions for these types of simulations
typically requires careful mesh construction. Conventional methods of mesh
generation involve either unstructured or structured (block) meshing. Structured
mesh generation involves generating conformal blocks within the flow field
containing grid size and spacing information from which the resulting fluid mesh
can be defined. For complicated geometries, however, this method of meshing is
intensive and can require weeks to generate a proper block mesh. Unstructured
meshing, on the other hand, does not require block definitions but instead
requires only the surface definition. This approach is faster but requires search
routines and other procedural checks within the meshing algorithm to prevent the
creation of malformed cells. The resulting volume mesh quality is dependent on
the surface mesh quality, and quite often several days are needed to prepare a
complex geometry with the proper surface definition necessary for meshing. Most
commercial CFD tools utilize unstructured mesh generation.
Both structured and unstructured meshing techniques are body-fitted grid
methods; there exists a third option, however, using the immersed boundary
methods which were discussed in 4.3.2. The intricacies of generating a grid
become less important and the burden of resolving the interface between the
fluid and solid bodies is shifted to the discretization algorithm. One benefit of this
technique is elimination of malformed or skewed cells; the body can be immersed
in a uniform isotropic grid, or for problems of large domain size localized block
refinement can ensure proper cell sizes to resolve the flow quantities of interest.
This allows for standardization of a background fluid mesh with which varying
geometries can be quickly analyzed without the need to redefine the mesh for
each surface. Such an advantage has applications for fast prototyping in the
automotive and motorsports industries, where varying external vehicle
components such as side mirrors and downforce-generating wings may be

119
quickly analyzed to determine the effect on the general flow around the car. To
demonstrate the potential of VC as an implicit immersed boundary method to
model a complex body, a Formula One racecar is simulated with the VC model at
8 yaw in a 3 m high by 5 m wide wind tunnel environment.
5.5.1 Time-Accurate VC Study
The simulated domain with outer dimensions marked is shown in Figure
64 facing downstream in a perspective view. The F1 car is a 60% scale vehicle
which would typically be tested in a wind tunnel with the given test section
dimensions. The vehicle is placed with the front wheel base 2.5 meters from the
inlet of the domain. This provides ample distance between the immersed body
and the inlet to avoid undesired boundary interactions. The domain extends 12.5
meters downstream of the front wheel base to highlight the important wake
region behind the F1 car. The wake is typically the most difficult portion of the

Figure 64. 5.5 case study domain with outer boundary dimensions marked.

120
flow field to properly resolve with conventional methods due to diffusion and
spreading of the vortical features; it is expected that the application of VC in both
near and far wake regions will result in an improvement in wake resolution. The
detailed immersed vehicle is displayed more closely in Figure 65, and is based
on a 2010-2011 season F1 car. Resolving the fine details of the car, such as the
exact curvature of the front and rear wings and body or the thin rear deflector
plates, is important for estimating surface pressures but is of secondary
importance for the wake region. To represent the surface without a body-fitted
grid, the fine details are approximated by setting the velocity of any cells within
the body to zero; naturally this approach neglects some small details but the
overall effect of the surface is taken into account. Applying the VC model results
in a smooth vorticity sheet generated from the interface between the interior and
exterior cells.
The OpenFOAM utility blockMesh has been used to create a coarse initial
base Cartesian mesh which matches the dimensions of the domain in Figure 64.

Figure 65. Surface representation of the immersed F1 car.

121
This mesh has been locally refined in both the wake region and around the
immersed surface. The OpenFOAM meshing utility snappyHexMesh was altered
to perform this function; the modified tool identifies cells within refinement regions
as well as cells intersecting the immersed surface and applies 2:1 refinement to
specified cell size levels without removing the cells within the body necessary for
the immersed boundary algorithm. The resulting refined Cartesian mesh with 2:1
interfaces between cell sizes is shown in Figure 66 and contains 3,400,000 cells.
Constant refinement is used throughout most of the wake region to mitigate the
cell size variation sensitivity of the VC2 model suggested by the 5.4 case study
results. The cells near the surface of the vehicle are allowed to refine two levels
smaller than the wake region to better approximate the curvature of the body with
the immersed boundary algorithm.
The VC2 model parameters are set to = 3.0 with = 0.001. Freestream
velocity is set to 50 m/s. Both the inlet and the flow field are set to 50 m/s and the
time-accurate simulation is impulsively started with a maximum CFL number of
10. After running one second solution time (3.3x flow pass-through) to wash out
the impulsive start effects, the time step is halved for a maximum CFL number of
5 to enhance the temporal flow resolution and the simulation is further advanced
to achieve a two second effective solution time.
5.5.2 High Resolution RANS Study
The 5.4 case study demonstrated the ability of VC to resolve more
turbulent length scales on a coarser mesh than necessary for conventional
methods. To provide a high resolution baseline to assess the effectiveness of the
VC2 model to simulate detailed flow around the complex immersed body and
within the wake region, the VC2 solution is compared to a high resolution
30,000,000 cell Reynolds-averaged Navier-Stokes (RANS) simulation of the
same Formula One car performed with the commercial CFD solver STAR-CCM+.
The high resolution study is available from a Jacobs Technology internal report
investigating yaw effects within an adaptive wall wind tunnel configuration [81].

122

Figure 66. Cartesian mesh used for 5.5 case study containing 3,400,000 cells; zoomed images show the spanwise refinement
through the car (top left) and streamwise refinement along the centerline (top right). The full mesh (centerline slice) is pictured in
the bottom image.

123

Figure 67. Reference geometry for high resolution STAR-CCM+ study including upstream
contraction.
The domain for the high resolution study is shown in Figure 67, and
includes the upstream contraction section of the wind tunnel as well as the test
section. Besides the upstream contraction, the high resolution simulation also
includes rolling road and boundary layer removal system effects, rotating wheels,
and engine intake and exhaust. These additional elements are not included in the
VC simulation because the current implementation of the immersed boundary in
OpenFOAM provides only a no-slip wall condition. However, the objective of the
comparison is to show the effect of VC within the wake region and this can be
achieved without including the additional effects.
The mesh in the high resolution study is a body-fitted grid comprised of 30
million polyhedral cells with near-wall resolution sufficient to apply a realizable k-
turbulence model with a two-layer wall function. Further explanation of the origins
and merits of the k- turbulence model can be found in Wilcox [78] and will not be
discussed here. Images of the mesh have been included in Figure A-7 in the

124
Appendix for reference. It is important to note that the unstructured polyhedral
cells in STAR-CCM+ are constructed from an initial tetrahedral mesh; each
polyhedral cell is created by joining 6-8 tetrahedral cells, implying that the 30
million polyhedral cell mesh is equivalent to a 180-240 million tetrahedral mesh.
A mesh of this overall size is required by conventional CFD methods to attempt
to resolve near-body and wake effects. Using VC, however, the resolution of
smaller scale behavior is possible on a much coarser mesh than required by
conventional methods.
5.5.3 Vorticity Field Comparison
An instantaneous snapshot from VISUALIZATION 17 of the vorticity in the
wake of the vehicle at the 3.0 second time step is presented for the time-accurate
VC simulation in an isometric view in Figure 68. The vorticity has been visualized
with a ray-tracing technique allowing all contour levels to be viewed
simultaneously. This provides a high degree of detail compared to an iso-contour
which allows for only one contour level to be viewed. In the visualization, higher
magnitudes of vorticity are represented by green and light blue shades, while
lower levels are dark blue. Vorticity leaves the immersed surface of the vehicle in
sheets which separate into smaller vortical structures within the wake as they
convect downstream. The tip vortices from the front wing appear to be resolved
through the full length of the domain and extend away from the vehicle on either
side. Wing tip vortices are exceedingly difficult to capture with conventional
modeling techniques and yet play a significant role in the determination of wall
pressures for an adaptive wall wind tunnel [81].
To compare the VC solution directly to the high resolution RANS study, an
iso-contour of vorticity at a single arbitrary magnitude of 150 is visualized for both
simulations. The wake refinement for both cases is comparable or even slightly
finer with the high resolution case which allows for an accurate and meaningful
comparison using this approach. A front perspective view of the instantaneous
vorticity iso-contour from the VC solution is shown in Figure 69.

125

Figure 68. Instantaneous snapshot of vorticity magnitude from VC2 solution at 3.0 s visualized with ray-tracing technique.

126

Figure 69. Front perspective view of vorticity magnitude iso-contour at a magnitude of 150
for time-accurate VC solution.

127
A chaotic mix of both large and small scale structures results from the
immersed surface and is convected downstream; rather than dissipating, the
structures are sustained by the VC model within the uniformly refined wake
region. A perspective view of the same iso-contour value is shown for the steady-
state RANS simulation in Figure 70. The flow field in the wake region is much
more subdued than seen in the VC simulation, resolving only large scale
structures shed from points of separation which dissipate shortly downstream.
Typically, steady-state simulations are used to resolve the time-averaged
behavior of the flow; however, the true time-averaged behavior of the turbulent
wake structures seen in the VC simulation would result in a much more
dispersed surface than provided by the steady-state iso-contour.
Additional views of both iso-contours from the side of the domain are
provided in Figure 71 and Figure 72, and are overlaid against the background
mesh in Figure 73. The steady-state RANS result indicates that the vortical
structures dissipate fully after approximately two car lengths downstream, while
the time-accurate VC solution shows structures traveling the full distance
downstream through the domain until a larger mesh size is encountered. The
abrupt truncation of the structures occurs because the effect of the VC term
changes with the larger cell downstream, and the absolute value of vorticity
magnitude decreases with increasing cell size. Nonetheless, with VC the vortical
structures are maintained and enhanced for more than three car lengths.
5.5.4 Streamline Comparison
For a final comparison between the VC and RANS simulations, streamline
visualizations are displayed in Figure 74 and Figure 75. The streamlines from the
VC solution are based on the time-averaged velocity over two second duration,
while the RANS streamlines are based on the steady-state velocity field. The key
difference between the two streamline plots is the structure of the wake region.
The RANS solution suggests the presence of a tight vortical structure located
directly behind the rear wing assembly which extends downstream, while the VC

128

Figure 70. Front perspective view of vorticity magnitude iso-contour at a magnitude of 150
for steady-state high resolution RANS solution.

129

Figure 71. Side view of vorticity magnitude iso-contour at a magnitude of 150 for the time-accurate VC solution.

Figure 72. Side view of vorticity magnitude iso-contour at a magnitude of 150 for the steady-state high resolution RANS
solution.

130

Figure 73. Comparison between vorticity magnitude iso-contours from the VC solution (top) and the high resolution RANS
solution (bottom) for the turbulent wake region with mesh refinement visible.

131

Figure 74. Streamline visualization based on time-averaged velocity for time-accurate VC simulation, colored by velocity
magnitude.

132

Figure 75. Streamline visualization of steady-state RANS solution, colored by velocity magnitude.

133
solution resolves a larger vehicle-sized vortical structure which slowly twists
downstream. The RANS solution appears to overemphasize the singular vortical
nature of the wake, whereas the wake structure in the VC simulation appears
more realistic. Furthermore, the VC solution suggests a set of streamlines turning
towards the right side wall (oriented looking downstream). This behavior is not
evident in the streamlines from the RANS simulation which indicate full
downstream flow directed towards the centerline of the vehicle. Wind tunnel
studies of Formula One vehicles at yaw have previously reported interactions
between the wall and tip vortices originating from the front wing [82], further
supporting the streamline directions observed from the VC solution.
The overall lack of resolved structures in the wake region in the RANS
simulation highlights the deficiencies of RANS turbulence modeling to resolve the
turbulent wake structures; even if the simulation was unsteady, the RANS model
would most likely not resolve the structures to the desired fidelity. For turbulent
wake simulation the best conventional modeling approach is large eddy
simulation. VC has been shown to exhibit qualities similar to an implicit LES
method. However, unlike LES simulations VC does not carry such stringent mesh
refinement requirements for vortical structure resolution and can still provide a
realistic solution on a coarser mesh.
Using a coarser mesh with VC provides several key benefits over a high
resolution mesh required in a RANS application. Overall mesh size is directly
proportional to hardware requirements as well as solution time; as a result, the
ability to solve a problem with less computational cost represents a significant
advantage. In this study, the high resolution simulation on 30 million cells
required two days to reach steady-state convergence on 46 processors, while the
time-accurate VC simulation with 3.4 million cells achieved a 2.3 second effective
solution duration in 18 hours on an equivalent number of processors. Not only
does the VC solution represent a savings in computational cost and schedule, it
also allows the vortical structures in the wake region to be simulated more
accurately than with a steady-state RANS turbulence model.

134
CHAPTER 6
CONCLUSION

In this thesis a set of parallel-computing, three-dimensional, fully functional
Navier Stokes solvers with Vorticity Confinement have been developed in C++
using the OpenFOAM computational framework. The solvers are comprised of
two separate algorithms for steady-state and parallel simulation, and because
they are based on the OpenFOAM framework they can easily interact with all of
the computational utilities and tools offered by OpenFOAM. Both VC1 and VC2
Vorticity Confinement formulations have been implemented and studied through
simulation. Additionally, a simple immersed boundary method has been
developed to be used specifically with the VC model.
The verification cases studied in Chapter 5 indicate that both formulations
of VC act to enhance the vortical features within the flow. The effect of VC on
boundary layers has been investigated with steady flow over a flat plate. The
results with both VC formulations indicate that for a fixed value of and a given
range of the thickness of the boundary layer can be controlled to a desired
height which approximates the shape of a turbulent boundary layer. Without any
outside perturbation the boundary layer remains at the desired height along the
boundary but can allow separation as necessary in response to outside
influences such as an adverse pressure gradient. As a formality the simulations
also verify the parallel implementation of the algorithm and exhibit solution
uniformity across processor partition boundaries. An additional boundary layer
case study demonstrates the implemented immersed boundary method with
simulations of an immersed slanted plate in a steady flow. Through studies
varying the numerical accuracy of the convection term, the effect of the VC2
formulation for a given set of and values is shown to be dependent on the
inherent numerical accuracy of the solver which confirms the findings of
Butsuntorn and Jameson [32]. The use of a second-order or higher numerical

135
scheme lowers the total numerical diffusion which must be taken into account
when using a set of and values for different simulations.
The solitary wave characteristic of VC is demonstrated with the time-
accurate advection of a compact vortex through a periodic domain. This case
study also provides an opportunity to observe the conservative aspects of both
VC formulations. The baseline simulation without VC illustrates that the compact
vortex dissipates before traveling one meter from the starting location. With both
VC formulations, the compact vortex is treated as a solitary wave and convected
with the background flow without diffusing or losing definition. The VC model
effectively behaves as a non-zero full asymptotic solution within the local region
of the vortex, preventing further dissipation. VC1 is verified to be non-
conservative of momentum, such that the streamwise motion of the vortex lags
and does not complete the full intended transit path within the simulated time.
The VC2 formulation completes the full transit, verifying its conservative
properties.
The first three-dimensional case study analyzes a backward-facing step to
provide a comparison of the VC model to experimental data. The performance of
the VC1 formulation for this problem is found to be invariant with respect to the
value of . For the VC2 formulation the results are shown to be heavily
dependent on the proportion of to . The reattachment length predictions are
increasingly poor with higher values; interrogating the flow field shows that with
higher values the shear layer mixes too quickly compared to the experiment
and the boundary layer is over-accelerated. This behavior appears to be the
result of enhancement of both larger and smaller eddies at higher values. The
VC1 and VC2 solutions show otherwise good correlation to both experimental
and LES data, highlighting the potential of VC to act as an implicit LES method.
Furthermore, as a turbulent flow qualifier the TKE spectra as a function of
wavenumber from the VC solutions indicate an inertial region with reference
slope of -5/3 matching the Kolmogorov hypothesis. With increasing a tendency
of the inertial wavenumber range to broaden over a larger span of wavenumbers

136
is observed, which is indicative of higher Reynolds number effects. An additional
set of simulations is analyzed with VC2 on a coarser mesh to indicate whether
the VC model is capable of providing identical quality of solution on coarser
meshes. Initial results with identical value from the previous simulations
indicate a thicker boundary layer similar to a lower Reynolds number solution.
The behavior suggests that on a coarser mesh the value of should be
decreased relative to an established value for a finer mesh to account for the
increased inherent numerical dissipation on the coarser mesh. The VC2
formulation with half the original value of provides a match to the experimental
data on the coarser mesh while the laminar solution does not correlate well to the
experimental flow field. This suggests that VC is able to resolve higher Reynolds
number turbulent phenomena on a coarser mesh than necessary for
conventional methods.
The final case study is a demonstration of the VC model with flow over a
complex body. A Formula One racecar is modeled in a 3 x 5 meter wind tunnel
domain at an 8 degree yaw configuration. A time-accurate VC solution using VC2
with the developed solver on a 3.4 million cell model is computed on 47
processors for 18 hours resulting in a 3.0 second solution time. This case is
compared to a steady-state RANS solution using the commercial CFD solver
STAR-CCM+ on a 30 million cell model computed on an equivalent number of
processors for 48 hours to achieve convergence. With the VC model, the vortical
structures in the wake region are resolved and preserved for more than three car
lengths while using the RANS model they dissipate entirely within 1.5 car
lengths. Streamline plots suggest that the VC solution clearly shows the effect of
the front wing tip vortex in the downstream wake region, while the steady-state
RANS simulation exhibits no such effect. The comparison demonstrates the
deficiencies of the steady-state RANS turbulence model to accurately resolve the
wake flow features, even with a high resolution mesh, and suggests that VC is
able to preserve the vortical structures in the wake on the coarser mesh.

137
6.1 Future Work
The current study has identified a range of acceptable confinement
parameters within the same order of magnitude which exhibit comparable
behavior for several non-similar simulations. Nonetheless, resolution of specific
effects in each case requires minor tuning on the parameters within the identified
range. Additional studies are needed to help refine and standardize the
confinement parameters to acceptable ranges of values for specific flow regimes
before using the developed set of solvers as a production CFD code. This
process of calibration is expected for the development of any turbulence model,
and requires executing more verification cases. Currently the specification of the
confinement parameters is somewhat ad hoc on a case-by-case basis;
establishing a well-defined range of acceptable parameters will help formalize the
method as a potential tool for turbulent simulation. Based on the results of the 5.4
case study indicating higher Reynolds number effects, parameterization of the
confinement parameters versus Reynolds number may be a possibility.
Several areas also exist for improvement in both the VC formulation itself
and the developed flow solver in OpenFOAM. The results of the immersed
slanted boundary case obviously confirmed that the effects of the VC model are
dependent implicitly on the numerical diffusion and as a result vary with
discretization scheme. Within OpenFOAM, to measure numerical error there
exist two local error estimate methods. The Moment Error Estimate derives the
solution error from cell imbalances based on higher moments of the solution
equations. The Residual Error Estimate, on the other hand, calculates error
based on local inconsistencies between face interpolation and volume integration
[48]. Studies are required to formulate an explicit dependence of on the local
solution error which would potentially help mitigate variations of the VC model
between discretization schemes. This type of enhancement would also be locally
cell dependent, leading to a dynamic specification of the VC parameter.
The transient effects in the boundary layer of the VC2 formulation in the
5.2 case study also show a potential for the development of a VC-based

138
boundary layer model which accurately models turbulence behavior near the
boundary. Real turbulent boundary layers are characterized by a multitude of
rolling vortical structures within the boundary layer, similar to the demonstrated
VC2 results. More study is needed to characterize the effects of VC near the
boundary and develop a predictive model for an accurate boundary layer.
More studies are also needed to develop a sufficient scaling parameter for
non-uniform meshes. The implemented cell size scaling method in this thesis
demonstrated questionable results in the downstream portion of the 5.4 case
studies where the mesh exhibited higher aspect ratio differences between cells.
A sufficient scaling method should not be detrimental and should also help
prevent variations of the VC effect across differences in cell size. However, such
a scaling method should also not be considered a panacea for variations in mesh
density; near-isotropic cell shapes have been verified to work properly with VC
and thus meshes for a VC simulation should be designed with certain cell aspect
ratio limitations in mind.

139

BIBLIOGRAPHY

140
BIBLIOGRAPHY

[1] J. H. Ferziger and M. Peric, Computational Methods for Fluid Dynamics, 3rd
ed. New York, NY: Springer-Verlag, 2001.
[2] John Steinhoff, "Vorticity Confinement: A New Technique for Computing
Vortex Dominated Flows," Frontiers of Computational Fluid Dynamics, pp.
236-264, 1994.
[3] John Steinhoff and David Underhill, "Modification of the Euler equations for
"vorticity confinement": Application to the computation of interacting vortex
rings," Physics of Fluids, vol. 6, pp. 2738-2744, 1994.
[4] John Steinhoff, Yonghu Wenren, and Lesong Wang, "Efficient Computation
of Separating High Reynolds Number Incompressible Flows Using Vorticity
Confinement," in 14th AIAA Computational Fluid Dynamics Conference,
AIAA-97-3316, Norfolk, VA, 1999.
[5] Meng Fan, W. Dietz, Yonghu Wenren, and John Steinhoff, "Computing
Complex Flows on Coarse Grids Using Vorticity Confinement," in 40th AIAA
Aerospace Sciences Meeting and Exhibit, AIAA-2002-0135, Reno, NV, 2002.
[6] Meng Fan and John Steinhoff, "Computation of Blunt Body Wake Flow By
Vorticity Confinement," in 42nd AIAA Aerospace Sciences Meeting and
Exhibit, AIAA-2004-0592, Reno, NV, 2004.
[7] John Steinhoff and Nicholas Lynn, "Treatment of Vortical Flow Using
Vorticity Confinement," Frontiers of Computational Fluid Dynamics, pp. 199-
240, 2006.
[8] Mitsuhiro Murayama, Kazuhiro Nakahashi, and Shigeru Obayashi,
"Numerical Simulation of Vortical Flows Using Vorticity Confinement Coupled
With Unstructured Grid," in 39th Aerospace Sciences Meeting and Exhibit,
AIAA-2001-0606, Reno, NV, 2001.
[9] M. Malek Jafarian and M. Pasandideh Fard, "Development and Application
of Compressible Vorticity Confinement," Scientia Iranica, vol. 14, no. 3, pp.
251-262, 2007.

141
[10] M. Forman, H. Jasak, J. Volavy, M. Jicha, and C. Othmer, "Vorticity
Confinement method applied to flow around an Ahmed body and comparison
with experiments," in Advances in Fluid Mechanics VIII, M. Rahman and C.
A. Brebbia, Eds.: WIT Press Publishing, 2010, vol. 69, pp. 55-66.
[11] Hrvoje Jasak, OpenFOAM: Introduction and Overview, March 31, 2009.
[12] John Steinhoff and K. Ramachandran, "Free Wake Analysis of Helicopter
Rotor Blades in Hover Using a Finite Volume Technique," DAAG29-84-K-
0019 1988.
[13] K. Ramachandran, John Steinhoff, and Yonghu Wenren, "Free Wake
Computation of Helicoptor Rotor Flowfield for General Flight Regimes,"
DAAL03-89-C-0027 1992.
[14] K. Ramachandran, S. Schlectriem, F. X. Caradonna, and John Steinhoff,
"The Application of Vorticity Embedding to the Computation of Advancing
Rotor Flows," in American Helicopter Society 49th Annual Forum, St. Louis,
MO, 1993.
[15] John Steinhoff, Nicholas Lynn, and Lesong Wang, "Large Eddy Simulation
Using Vorticity Confinement," in Implicit Large Eddy Simulation: Computing
Turbulent Flow Dynamics, F. F. Grinstein, L. G. Margolin, and W. J. Rider,
Eds. New York, NY: Cambridge University Press, 2007, ch. 4e, pp. 168-195.
[16] John Steinhoff, Meng Fan, Lesong Wang, and William Dietz, "Convection of
Concentrated Vortices and Passive Scalars as Solitary Waves," Journal of
Scientific Computing, vol. 19, no. 1-3, pp. 457-478, 2003.
[17] Yonghu Wenren, John Steinhoff, Lesong Wang, Meng Fan, and Min Xiao,
"Application of Vorticity Confinement to the Prediction of the Flow over
Complex Bodies," in Fluids 2000 Conference and Exhibit, AIAA-2000-2621,
Denver, CO, 2000.
[18] John Steinhoff et al., "Turbulent Flow Simulations Using Vorticity
Confinement," in Implicit Large Eddy Simulation: Computing Turbulent Flow
Dynamics, F. F. Grinstein, L. G. Margolin, and W. J. Rider, Eds. New York,
NY: Cambridge University Press, 2007, ch. 12, pp. 370-392.
[19] Jeong-Mo Hong, Ho-Young Lee, Jong-Chul Yoon, and Chang-Hun Kim,
"Bubbles Alive," ACM Transactions on Graphics, vol. 27, no. 3, 2008.

142
[20] Taekwon Jang, Heeyoung Kim, Jinhyuk Bae, Jaewoo Seo, and Junyong
Noh, "Multilevel vorticity confinement for water turbulence simulation," The
Visual Computer: International Journal of Computer Graphics, vol. 26, no. 6-
8, pp. 873-881, 2010.
[21] Timothy Suttles, D. Brian Landrum, Brian E. Greiner, and Michael A.
Robinson, "Calibration of Vorticity Confinement Techniques for Missile
Aerodynamics: Part I - Surface Confinement," in 42nd AIAA Aerospace
Sciences Meeting, AIAA-2004-719, Reno, NV, 2004.
[22] Milton E. Vaughn, "An Application of Vorticity Confinement to Missile
Aerodynamic Design," in 24th Applied Aerodynamics Conference, AIAA-
2006-3866, San Francisco, CA, 2006.
[23] John Steinhoff, E. Puskas, S. Babu, Y. Wenren, and David Underhill,
"Computation of Thin Features Over Long Distances Using Solitary Waves,"
in 13th AIAA Computational Fluid Dynamics Conference, AIAA-97-1976,
Snowmass Village, CO, 1997.
[24] Nicholas F. Lynn. (2007) Investigation of Vorticity Confinement as a High
Reynolds Number Turbulence Model. [Online]. In TRACE,
http://trace.tennessee.edu/utk_graddiss/235
[25] M. Costes and F. Juillet, "Analysis and higher-order extension of the VC2
confinement scheme," Computers & Fluids, December 2011. [Online].
http://www.sciencedirect.com/science/article/pii/S0045793011003653
[26] Guangchu Hu, The Development and Applications of a Numerical Method for
Compressible Vorticity Confinement in Vortex-Dominant Flows, 2001.
[27] Guangchu Hu and Bernard Grossman, "The computation of massively
separated flows using compressible vorticity confinement methods,"
Computers & Fluids, vol. 35, pp. 781-789, 2006.
[28] Michel Costes, "Analysis of the second vorticity confinement scheme,"
Aerospace Science and Technology, vol. 12, pp. 203-213, 2008.
[29] Ronald Fedkiw, Jos Stam, and Henrik Wann Jensen, "Visual Simulation of
Smoke," in SIGGRAPH 2001, 2001, pp. 23-30.
[30] Rainald Lhner, Chi Yang, and Robert Roger, "Tracking Vortices Over Large
Distances Using Vorticity Confinement," in 24th Symposium on Naval
Hydrodynamics, Fukuoka, Japan, 2002.

143
[31] Michael Robinson, "Application of Vorticity Confinement to Inviscid Missile
Force and Moment Prediction," in 42nd AIAA Aerospace Sciences Meeting,
AIAA-2004-0717, Reno, NV, 2004.
[32] Nawee Butsuntorn and Antony Jameson, "Time Spectral Method for
Rotorcraft Flow with Vorticity Confinement," in 26th AIAA Applied
Aerodynamics Conference, AIAA-2008-7340, Honolulu, Hawaii, 2008.
[33] Nawee Butsuntorn, "Time Spectral Method for Rotorcraft Flow with Vorticity
Confinement," PhD dissertation, 2008.
[34] Michel Costes and Gunawan Kowani, "An automatic anti-diffusion method
for vortical flows based on Vorticity Confinement," Aerospace Science and
Technology, vol. 7, pp. 11-21, 2003.
[35] Seonghyeon Hahn and Gianluca Iaccarino, "Towards Adaptive Vorticity
Confinement," in 47th AIAA Aerospace Sciences Meeting, AIAA-2009-1613,
Orlando, FL, 2009.
[36] H. G. Weller, G. Tabor, H. Jasak, and C. Fureby, "A tensorial approach to
computational continuum mechanics using object-oriented techniques,"
Computers in Physics, vol. 12, no. 6, pp. 620-631, 1998.
[37] Eric Paterson, "OverFOAM," in 9th Symposium on Overset Composite Grids
and Solution Technology, State College, PA, 2008.
[38] Silicon Graphics International Corp. (2011) OpenFOAM: The open source
CFD toolbox. [Online]. http://www.openfoam.com
[39] The Extend Project. (2011) Home - The OpenFOAM Extend Project.
[Online]. http://www.extend-project.de/
[40] Thomas Anderson. (2008, August) NACHOS C++ introduction. [Online].
http://www.cs.washington.edu/homes/tom/c++example/c++.pdf
[41] Bjarne Stroustrup, "An Overview of the C++ Programming Language," in The
Handbook of Object Technology, Saba Zamir, Ed. Boca Raton: CRC Press
LLC, 1999, ch. 15.
[42] Bjarne Stroustrup, The C++ Programming Language: Special Edition, 3rd ed.
Indianapolis, IN: Addison-Wesley Professional, 1997.

144
[43] Hrvoje Jasak, Henry Weller, and Niklas Nordin, "In-Cylinder CFD Simulation
Using a C++ Object-Oriented Toolkit," in SAE 2004 World Congress &
Exhibition, SAE-2004-01-0110, Detroit, MI, 2004.
[44] OpenFOAM Foundation. (2011, December) OpenFOAM User Guide v2.1.0.
[45] Juha Kortelainen, Paul Klinge, and Kai Katajamki, "Managing Mesh-Based
Data in a Semantic Database," Helsinki, Finland, Research Report VTT-R-
03154-11 2011.
[46] Silicon Graphics International Corp. (2011) Run-time Post-processing.
[Online]. http://www.openfoam.com/features/runtime-postprocessing.php
[47] Bernard Gschaider. (2012, January) Unofficial OpenFOAM Wiki: Main Page.
[Online]. http://openfoamwiki.net/index.php/Main_Page
[48] Hrvoje Jasak, "Error Analysis and Estimation for the Finite Volume Method
with Applications to Fluid Flows," PhD dissertation, 1996.
[49] John Tannehill, Dale Anderson, and Richard Pletcher, Computational Fluid
Mechanics and Heat Transfer, 2nd ed. Washington, DC: Taylor & Francis,
1997.
[50] John Anderson, Computational Fluid Dynamics: The Basics with
Applications. New York, NY: McGraw-Hill, Inc., 1995.
[51] A. J. Chorin, "Numerical solution of the Navier-Stokes equations,"
Mathematics of Computation, no. 22, pp. 745762, 1968.
[52] Dimitris Drikakis and William Rider, High-Resolution Methods for
Incompressible and Low Speed Flows. Berlin: Springer-Verlag, 2005.
[53] Raad I. Issa, "Solution of the Implicitly Discretised Fluid Flow Equations by
Operator Splitting," Journal of Computational Physics, vol. 62, pp. 40-65,
1985.
[54] Raad I. Issa, A. D. Gosman, and A. P. Watkins, "The Computation of
Compressible and Incompressible Recirculating Flows by a Non-iterative
Implicit Scheme," Journal of Computational Physics, vol. 62, pp. 66-82,
1986.
[55] Suhas V. Patankar, Numerical Heat Transfer and Fluid Flow. Washington,
D.C.: Hemisphere, 1980.

145
[56] C. M. Rhie and W. L. Chow, "Numerical study of the turbulent flow past an
airfoil with trailing edge separation," AIAA Journal, vol. 21, pp. 1525-1532,
1983.
[57] Rajat Mittal and Gianluca Iaccarino, "Immersed Boundary Methods," Annual
Review of Fluid Mechanics, vol. 37, pp. 239-261, 2005.
[58] C. S. Peskin, "The Fluid Dynamics of Heart Valves: Experimental,
Theoretical, and Computational Methods," Annual Review of Fluid
Mechanics, vol. 14, pp. 235-259, 1982.
[59] Ming-Chih Lai and C. S. Peskin, "An Immersed Boundary Method with
Formal Second-order Accuracy and Reduced Numerical Viscosity," Journal
of Computational Physics, vol. 160, no. 2, pp. 705-719, 2000.
[60] J. Mohd-Yosuf, "Development of Immersed Boundary Methods for Complex
Geometries," Annual Research Briefs 1998.
[61] Yu-Heng Tseng and Joel Ferziger, "A ghost-cell immersed boundary method
for flow in complex geometry," Journal of Computational Physics, vol. 192,
pp. 593-623, 2003.
[62] R. Ghias, R. Mittal, and H. Dong, "A sharp interface immersed boundary
method for compressible viscous flows," Journal of Computational Physics,
vol. 225, pp. 528-553, 2007.
[63] Andreas Mark and Berend van Wachem, "Derivation and validation of a
novel implicit second-order accurate immersed boundary method," Journal of
Computational Physics, vol. 227, pp. 6660-6680, 2008.
[64] H. S. Udaykumar, R. Mittal, P. Rampunggoon, and A. Khanna, "A Sharp
Interface Cartesian Grid Method for Simulating Flows with Complex Moving
Boundaries," Journal of Computational Physics, vol. 174, pp. 345-380, 2001.
[65] AIAA, "Guide for the Verification and Validation of Computational Fluid
Dynamics Simulations," Reston, VA, 1998.
[66] Personal communication with John Steinhoff, "Discussion on confinement
parameters", February 2012.
[67] John Steinhoff, Subhashini Chitta, and Paula Sanematsu, "Capturing Short
Waves Over Long Distances As Nonlinear Solitary Waves," in 7th IMACS
International Conference, Athens, GA, 2011.

146
[68] A. R. Bishop, J.A. Krumhansl, and S.E. Trullinger, "Solitons in Condensed
Matter: A Paradigm," Physica D: Nonlinear Phenomena, vol. 1, no. 1, pp. 1-
44, 1980.
[69] Hans J. Lugt, Introduction to Vortex Theory. Potomac, MD: Vortex Flow
Press, Inc., 1996.
[70] R. W. Pitz and J. W. Daily, "Combustion in a turbulent mixing layer formed at
a rearward-facing step," AIAA Journal, vol. 21, pp. 1565-1570, 1983.
[71] J. O. Keller, R. W. Pitz, J. L. Eilzey, I. G. Shepard, and J. W. Daily, "The
structure and dynamics of reacting plane mixing layers," Experiments in
Fluids, vol. 6, pp. 33-43, 1988.
[72] H. G. Weller, G. Tabor, A. D. Gosman, and C. Fureby, "Application of a
Flame-Wrinkling LES Combustion Model to a Turbulent Mixing Layer," in
27th Symposium (International) on Combustion, 1998, pp. 899-907.
[73] Nam Seob Park and Sang Cheol Ko, "Large eddy simulation of turbulent
premixed combustion flows over backward facing step," Journal of
Mechanical Science and Technology, vol. 25, no. 3, pp. 713-719, 2011.
[74] Pierre Sagaut, Large Eddy Simulation for Incompressible Flows: An
Introduction, 3rd ed. Heidelberg, Germany: Springer-Verlag Berlin, 2006.
[75] B. F. Armaly, F. Durst, J. C. F. Pereira, and B. Schoenung, "Experimental
and theoretical investigation of backward-facing step flow," Journal of Fluid
Mechanics, vol. 127, pp. 473-496, 1983.
[76] N. Nait Bouda, R. Schiestel, M. Amielh, C. Rey, and T. Benabid,
"Experimental approach and numerical prediction of a turbulent wall jet over
a backward facing step," International Journal of Heat and Fluid Flow, vol.
29, pp. 927-944, 2008.
[77] S. B. Pope, Turbulent Flows. Cambridge, UK: Cambridge University Press,
2000.
[78] David C. Wilcox, Turbulence Modeling for CFD. La Caada, CA: DCW
Industries, Inc., 1994.
[79] Austin Kimbrell, Joseph Yen, and Christopher Connor, "CAA Study of WICS
Data Using JUSTUS," in ITEA 15th Aircraft / Stores Compatibility
Symposium, Fort Walton Beach, FL, 2010.

147
[80] Andrew Aspden, Nikos Nikiforakis, Stuart Dalziel, and John B. Bell, "Analysis
of Implicit LES methods," Communications in Applied Mathematics and
Computational Science, vol. 3, no. 1, pp. 103-126, 2008.
[81] Jacobs Technology, Inc., "Effects of Vehicle at Yaw on Adaptive Wall Wind
Tunnel Shape," Internal report, 2011.
[82] Personal communication with Bill Martindale, "Wall effects of wing tip
vortices", March 23, 2012.

148

APPENDIX

149

Figure A-1. Iso-volume of mean streamwise velocity U < 0.0 for VC1 case, 5.4 case study.

Figure A-2. Iso-volume of mean streamwise velocity U < 0.0 for VC2 case, 5.4 case study
with = 1.6.

150

with = 2.0.

with = 2.5.

151

Figure A-5. TKE spectrum versus wavenumber for VC1 solution, 5.4 case study.

152

Figure A-6. TKE spectrum versus wavenumber for VC2 = 2.0 solution, 5.4 case study.

153

Figure A-7. Mesh from high resolution RANS study for 5.5 case study containing 30 million polyhedral cells; zoomed images
show the local refinement around the car (top left) and streamwise refinement along the centerline (top right). The section of the
mesh (centerline slice) corresponding to the VC2 simulation domain dimensions is pictured in the bottom image.

154
VITA

Austin Barrett Kimbrell was born in Hendersonville, TN on August 26
th
,
1985, to the parents of Donald and Deanna Kimbrell. He was homeschooled
through primary school and later attended Hendersonville High School,
graduating salutatorian in 2003.

After graduation, Austin studied both engineering and art at the University of
Tennessee in Knoxville. He completed a year in a German language study
abroad program as an exchange student at Phillipps Universitt in 2005 where
he met his wife, Oxana. Austin returned to the US and graduated summa cum
laude from the University of Tennessee in December 2008 with a Bachelor of
Science in Mechanical Engineering and a minor in materials science.

Austin accepted a position at Jacobs Technology as a computational fluids
engineer and aerodynamicist after graduation. In the fall of 2009, he began
studying at the University of Tennessee Space Institute as a part-time graduate
student. He graduated with a Masters of Science in Mechanical Engineering in
May 2012.

Thesis Used Openfoam

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Thesis Used Openfoam

Uploaded by

Copyright:

Available Formats

University of Tennessee, Knoxville

Trace: Tennessee Research and Creative

term in (2.7) can be interpreted as the normalized vorticity vector

term in (2.7). The harmonic mean is

. Each control volume is associated with a coefficient

. The solution requires the

and off-diagonal coefficients

, the U vector matrix has one

and H(U) terms in Equation

You might also like